Real-time nondestructive testing system for turbine blade
By printing or embedding microelectrodes on the surface of the turbine blades, combining the current collector module and deep neural network model, the comprehensive and real-time problems of turbine blade detection are solved, and high-accurate non-destructive detection is achieved.
Patent Information
- Application Number
- CN202510459086.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-14
- Publication Date
- 2025-07-18
AI Technical Summary
The existing turbine blade detection methods cannot achieve full coverage and poor resolution, and cannot be detected during the high-speed rotation of the blade, affecting the turbine working efficiency.
The micro-electrodes are printed on the screen or embedded in the surface of the turbine blade, combined with the current collector module, preprocessing module, identification module and alarm module, and the resistance distribution information is obtained in real time through current data, and defect identification is used to use the deep neural network model to identify defects, and detect them during high-speed rotation.
The comprehensive coverage, real-time and high-accuracy detection of turbine blades is achieved. There are no detection blind spots and are not affected by turbine blade materials, which improves the accuracy and efficiency of detection.
Smart Images
Figure CN120334297A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of turbine blade detection, and particularly to a real-time non-destructive detection system for turbine blades. Background Art
[0002] Currently, common non-destructive testing methods for turbine blades include dye penetrant inspection, ultrasonic surface wave testing, etc. However, dye penetrant inspection and ultrasonic testing are difficult to cover the entire surface of the blade. At the same time, the resolution is limited by the material structure, and it is difficult to monitor the high-speed rotating blade in real time. It can only perform fixed-point or intermittent detection when the blade stops rotating, which affects the continuous operation of the turbine.
[0003] Therefore, the technical problem solved by this application is that the existing turbine blade detection methods have insufficient coverage, poor resolution, and cannot detect the blade during its high-speed rotation, affecting the working efficiency of the turbine. Summary of the Invention
[0004] This application provides a real-time non-destructive detection system for turbine blades, which can achieve comprehensive, real-time, and highly accurate detection of turbine blades, without detection blind spots and not affected by the material of the turbine blades.
[0005] An embodiment of this application provides a real-time non-destructive detection system for turbine blades, including an electrode network, a current collection module, a preprocessing module, an identification module, and an alarm module; the electrode network includes a plurality of microelectrodes printed or embedded on the surface of the turbine blade to be measured with a preset density; the current collection module is connected to the electrode network through a brush and a conduction structure;
[0006] The current collection module is used to receive the current data of the electrode network when the turbine blade to be measured rotates; and, based on the current data and the electrode distribution information of the electrode network, obtain the resistance distribution information of the turbine blade to be measured; send the resistance distribution information to the preprocessing module;
[0007] The preprocessing module is used to preprocess the resistance distribution information and then send it to the identification module;
[0008] The identification module is used to input the preprocessed resistance distribution information into a defect identification model to obtain an identification result;
[0009] The alarm module is used to generate an alarm prompt message when the identification result meets a preset alarm type.
[0010] Furthermore, the system also includes a training module; the training module is used to obtain normal resistance distribution information of normal turbine blades under different working conditions and defect resistance distribution information of defective turbine blades with different defect types; and send the information to the preprocessing module; and obtain defect parameter information of each defective turbine blade; generate corresponding annotation data of defect resistance distribution information based on the defect parameter information; obtain a pre-training model, and train the pre-training model based on the annotation data, the received pre-processed normal resistance distribution information and the defect resistance distribution information to obtain a defect recognition model;
[0011] The preprocessing module is also used to preprocess the received normal resistance distribution information and defective resistance distribution information and then send them to the training module.
[0012] Furthermore, the current collecting module and the brushes on the rotor where the turbine blades to be tested are located are connected by sliding contact; the conductive structure extends along the surface of the rotor to the end of the rotor and is connected to the brushes at the end of the rotor.
[0013] Furthermore, the preprocessing module is specifically used to treat the resistance distribution information exceeding the preset threshold range as an abnormal value and eliminate it.
[0014] Furthermore, the preprocessing module is specifically used to determine missing data points in the resistance distribution information; generate interpolation data based on adjacent data of the missing data points and using a linear interpolation method or a mean filling method, and use the interpolation data as the missing data points.
[0015] Furthermore, the preprocessing module is specifically used to normalize the resistance distribution information.
[0016] Furthermore, the defect parameter information includes the crack length, crack depth, corrosion area and wear amount of the defective turbine blade.
[0017] Furthermore, the annotated data include blade geometry, defect type, severity, manufacturing method and acquisition time.
[0018] Furthermore, the pre-trained model is a deep neural network model, and the activation function is a ReLU function.
[0019] Furthermore, the alarm module is specifically used to output the alarm prompt information in the form of sound and light alarm, interface pop-up window or SMS push.
[0020] In summary, compared with the prior art, the technical solution provided in the embodiment of the present application has at least the following beneficial effects:
[0021] A real-time non-destructive testing system for turbine blades provided by an embodiment of the present application. First, the electrode network adopts printed integrated circuit technology to print or embed a large number of microelectrodes on the surface of the turbine blade to be tested, forming a structure that fully covers the blade, achieving comprehensive coverage detection of the turbine blade, having no detection blind spots and being unaffected by the material of the turbine blade. Second, the present application obtains the resistance distribution information through the current data of the electrode network during the high-speed rotation of the turbine blade, and performs defect detection on the turbine blade according to the resistance distribution information. It does not require the turbine blade to stop, and the recognition result obtained based on the resistance distribution information and the defect recognition model is more accurate than the existing detection methods, realizing real-time and highly accurate detection of the turbine blade. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 Schematic diagram of the connection of the electrode network, the current collection module, and the rotor provided by an exemplary embodiment of the present application
[0023] Figure 2 Schematic diagram of the arrangement of the electrode network on the turbine blade provided by an exemplary embodiment of the present application.
[0024] Figure 3 Structural diagram of a real-time non-destructive testing system for turbine blades provided by an exemplary embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0025] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments.
[0026] All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.
[0027] Please refer to Figure 1 、 Figure 2 and Figure 3 An embodiment of the present application provides a real-time non-destructive testing system for turbine blades, including an electrode network, a current collection module, a preprocessing module, an identification module, and an alarm module; the electrode network includes a plurality of microelectrodes printed or embedded on the surface of the turbine blade to be tested with a preset density distribution; the current collection module is connected to the electrode network through a brush and a conduction structure.
[0028] Among them, the preset density distribution of the electrode network needs to be closely adapted to the blade shape. According to the aerodynamic shape of the blade, such as contour features such as the leading edge, trailing edge, and blade body surface, the electrodes are evenly distributed and conform to the curvature of the blade surface to achieve the best detection effect.
[0029] The electrode density can be further appropriately increased in areas with stress concentration and prone to defects such as the blade tip and root. The blade tip is prone to defects such as cracks and wear due to the large air flow impact force and centrifugal force it bears; the root, due to its connection to the hub and complex stress, is a region where defects frequently occur. Increasing the electrode density can improve the detection sensitivity of these key parts, more accurately capture the resistance changes, timely detect potential defects, and enhance the detection ability for micro-defects. In addition, when designing the electrode distribution density, the stress conditions during the high-speed rotation of the blade can be fully considered. Ensure that the electrodes remain stable under complex loads such as centrifugal force and aerodynamic force, without loosening, falling off or being damaged, and ensure good contact between the electrodes and the blade surface, thereby ensuring the accuracy and stability of resistance measurement. By simulating the stress state during the high-speed rotation of the blade, the electrode layout is optimized so that the distribution of electrodes in areas with greater stress can meet the detection requirements and withstand the corresponding stress.
[0030] Specifically, the current collection module and the carbon brush on the rotor where the turbine blade to be measured is located are connected in a sliding contact manner; the conduction structure extends along the surface of the rotor to the rotor end and is connected to the carbon brush at the rotor end.
[0031] The current collection module can be regarded as a current collector head; the conduction structure extending from the blade electrode network has a smooth transition with the rotor surface to ensure stable transmission of signals and current during the high-speed rotation of the blade. This conduction structure is made of special materials and designs, has good electrical conductivity and mechanical strength, and can withstand the centrifugal force and vibration generated by high-speed rotation.
[0032] The conduction structure extends along the rotor surface to the rotor end and is connected to the carbon brush at the end. The carbon brush forms a sliding contact connection with the current collector head. This connection method allows the electrode network to maintain a stable electrical connection with the external power supply during the high-speed rotation of the rotor, realizing the supply of electrical energy and the transmission of resistance signals. The current collector head, as the gathering point of different regions of the electrode network, connects the electrode networks in various regions of the blade together. It plays a role in integrating resistance signals to ensure that the current data from the electrode network can be accurately and efficiently transmitted to the preprocessing module and the recognition module for analysis. The current collector head has good electrical conductivity and signal transmission performance, can reduce signal loss and interference, and ensure the accuracy and reliability of the measured potential resistance signals.
[0033] The current collection module is used to receive the current data of the electrode network when the turbine blade to be measured rotates; and, based on the current data and the electrode distribution information of the electrode network, obtain the resistance distribution information of the turbine blade to be measured; and send the resistance distribution information to the preprocessing module.
[0034] Specifically, the present application uses a brush to connect to an external power source, and the brush is in sliding contact with the collector head on the rotor to provide a stable power supply for the rotating electrode network. In the specific implementation process, only a limited voltage is given to the electrode network, and the minimum current is generated to obtain a resistance value, thereby reducing the impact on the turbine blades.
[0035] Furthermore, wireless oil-to-electricity technology can be used to realize wireless current power supply, which can be specifically achieved by inductive coupling or microwave emission; further, a partitioned electrode network can be made, and the electrode network in each area converges into a collector head, and then all the collector heads are connected to the pretreatment module through a spiral winding cable.
[0036] During the power supply process, according to Ohm's law (I = U / R), the resistance value at the corresponding position is calculated by measuring the current between each microelectrode or between the microelectrode and the reference point. Since the electrode network covers the surface of the blade, the resistance at different positions varies due to the material structure and defects of the blade, and the resistance distribution information can be obtained.
[0037] The preprocessing module is used to preprocess the resistance distribution information and then send it to the identification module.
[0038] The preprocessing module is specifically used to treat the resistance distribution information exceeding the preset threshold range as an abnormal value and remove it.
[0039] Specifically, the preprocessing module needs to remove outliers caused by sensor failure, electromagnetic interference, etc. during the acquisition process, and identify and delete data points that obviously deviate from the normal range by setting reasonable thresholds.
[0040] The preprocessing module is also specifically used to determine the missing data points in the resistance distribution information; generate interpolation data based on the adjacent data of the missing data points and using a linear interpolation method or a mean filling method, and use them as the missing data points.
[0041] Specifically, for missing data points, reasonable filling is performed based on the trends and correlations of adjacent data points, such as using linear interpolation or mean filling methods to ensure the continuity and integrity of the data.
[0042] The preprocessing module is also specifically used to normalize the resistance distribution information.
[0043] Specifically, the resistance data is normalized so that it is within a specific value range (such as between 0 and 1) to facilitate subsequent model training and calculation. Common normalization methods such as minimum-maximum normalization can be used here to convert each data point according to the minimum and maximum values of the collected resistance data.
[0044] The recognition module is used to input the preprocessed resistance distribution information into the defect recognition model to obtain the recognition result.
[0045] Specifically, by learning a large amount of historical normal and defective resistance change data, the resistance change characteristics are extracted to establish a defect recognition model, which can be compared with the real-time monitored resistance distribution information to improve the efficiency and accuracy of defect recognition.
[0046] It is worth noting that before the blade is put into use, a high-precision resistance measuring device can be used to conduct a comprehensive scan of the blade surface resistance under conditions similar to the actual working environment of the blade (considering factors such as temperature and humidity), and the scanned measurement data can be repeatedly measured and averaged multiple times to integrate these data into a comprehensive and accurate benchmark resistance data, so as to serve as the basis for the identification module to update the defect identification model. The benchmark resistance data not only contains resistance value data, but also records information such as measurement point coordinates and measurement environment parameters, which facilitates more accurate comparison and analysis, and generates more accurate identification results.
[0047] The alarm module is used to generate an alarm prompt message when the recognition result meets the preset alarm type.
[0048] Specifically, the recognition result may include a severity of mild, moderate or severe, and the preset alarm type may be set to moderate. When the severity of the recognition result is moderate or severe, the alarm module will generate an alarm prompt message.
[0049] In the specific implementation process, the alarm module is specifically used to output the alarm prompt information in the form of sound and light alarm, interface pop-up window or SMS push, or integrate with the enterprise monitoring platform, etc. As long as it ensures that the alarm information can be conveyed to relevant personnel in a timely and accurate manner, it can provide strong support for the safe operation of turbine blades.
[0050] A real-time nondestructive testing system for turbine blades is provided in the above-mentioned embodiment. Firstly, the electrode network adopts printed integrated circuit technology to print or embed a large number of micro-electrodes on the surface of the turbine blade to be tested, forming a structure that fully covers the blade, thereby realizing comprehensive coverage detection of the turbine blade, without any detection blind spots and without being affected by the turbine blade material; secondly, the present application obtains resistance distribution information through the current data of the electrode network during the high-speed rotation of the turbine blade, and performs defect detection of the turbine blade based on the resistance distribution information. There is no need to stop the turbine blade, and the recognition result obtained based on the resistance distribution information and the defect recognition model is more accurate than the existing detection method, thereby realizing real-time and high-accuracy turbine blade detection.
[0051] In some embodiments, the system further includes a training module; the training module is configured to obtain the normal resistance distribution information of normal turbine blades under different working conditions and the defective resistance distribution information of defective turbine blades of different defect types; and send them to the preprocessing module; and, obtain the defect parameter information of each defective turbine blade; generate labeled data corresponding to the defective resistance distribution information based on the defect parameter information; obtain a pre-trained model, and train the pre-trained model based on the labeled data, the received preprocessed normal resistance distribution information and defective resistance distribution information to obtain a defect recognition model.
[0052] First, collect the normal resistance distribution information of turbine blades operating normally under different working conditions (such as different rotational speeds, temperatures, pressures, etc.). These blades come from various models of turbines and have been strictly inspected to ensure no defects. For each blade, continuously collect the resistance distribution information at set time intervals (such as once per second) during its stable operating state. The collection duration covers the entire process from the start-up to the stable operation and then to the stop of the blade to obtain the comprehensive normal operating resistance change range.
[0053] Then, simulate different positions and degrees of various common defect types (such as cracks, corrosion, wear, inclusions, etc.) on the blades. Obtain defective turbine blades by artificially damaging the blades in the laboratory or from turbines with defects that have occurred during actual operation and have been confirmed by professional inspections. For each defect type, set different severity levels (such as minor, moderate, severe) and put them into the labeled data. Before and after manufacturing the defects, collect the resistance distribution information of the defective turbine blades respectively, and record the defect parameter information, where the defect parameter information includes the crack length, crack depth, corrosion area, and wear amount of the defective turbine blade. The labeled data includes blade geometry, defect type, severity, manufacturing method, and collection time. Compare the recognition result output by the model with the labeled data to update the model parameters.
[0054] Specifically, the pre-trained model can be a deep neural network (DNN) model, and the activation function is the ReLU function.
[0055] This application uses a deep neural network (DNN) as the basic architecture of the defect recognition model. The DNN has a powerful non-linear mapping ability and can learn the relationship between complex resistance changes and defects. The network structure includes multiple hidden layers, with each layer containing a certain number of neurons. For example, 3-5 hidden layers are adopted, and the number of neurons in each layer is adjusted according to the characteristics and complexity of the input data (such as 128, 256, 512, etc.). The ReLU (Rectified Linear Unit) function is selected as the activation function to introduce non-linearity in the hidden layer, accelerating the model training speed and improving the model performance. The output layer is designed according to the types and degrees of defects to be recognized. For example, a combination of multi-classification output (for different defect types) and regression output (for defect degrees) is adopted. The preprocessed resistance distribution information of normal and defective blades is divided into a training set and a validation set according to a certain ratio (such as 70% for training and 30% for validation). The DNN model is trained using the training set. The backpropagation algorithm is used to calculate the error between the recognition result and the labeled data, and the weights and biases of the model are continuously adjusted through optimization algorithms (such as stochastic gradient descent and its variants Adagrad, Adadelta, etc.) to minimize the error. During the training process, appropriate parameters such as the learning rate (such as 0.001), the number of iterations (such as 1000 times), and the batch size (such as 32) are set and adjusted according to the accuracy and loss values of the validation set to prevent overfitting. The performance of the model is regularly evaluated on the validation set, and indicators such as accuracy, recall rate, and F1 value are monitored. When the indicators no longer improve or signs of overfitting appear, the training is stopped.
[0056] After stopping the training, the trained model is comprehensively evaluated using an independent test set (extracted from actual operation data and not participating in the training and validation processes). Performance indicators such as the accuracy, precision, recall rate, and F1 value of the model on the test set are calculated to evaluate the model's ability to recognize different defect types and degrees.
[0057] Visualization tools such as confusion matrices are used to analyze the prediction results of the model, intuitively showing the correct recognition and misjudgment situations of the model for various defects, finding out the defect types that the model is prone to confuse or misjudge, and providing a basis for further optimizing the model.
[0058] According to the evaluation results, if performance bottlenecks or overfitting problems are found in the model, corresponding optimization measures are taken. For example, increase the amount of training data to further enrich the diversity of the data; adopt regularization techniques (such as L1 and L2 regularization) to prevent overfitting, and adjust the regularization parameters to balance the complexity and fitting ability of the model; adjust the model structure, such as increasing or decreasing the number of hidden layers, adjusting the number of neurons, etc., to optimize the model's expressive ability. At the same time, continuously introduce new actual operating data to update and optimize the model, so that the model can adapt to changes in different working conditions and blade states, and improve the generalization ability and accuracy of the model. Through the continuous evaluation and optimization process, ensure that the trained defect recognition model can accurately and reliably identify the types and degrees of defects on the surface of the turbine blade, providing an effective means of detection and early warning for the safe operation of the turbine blade.
[0059] The preprocessing module is also used to preprocess the received normal resistance distribution information and defect resistance distribution information and then send them to the training module. The preprocessing operations of the preprocessing module on the normal resistance distribution information and the defect resistance distribution information can refer to the operation process of the resistance distribution information in the above embodiments, which will not be elaborated here.
[0060] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.
[0061] The above-described embodiments only represent several implementation manners of the present application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the appended claims.
Claims
1. A real-time non-destructive testing system for turbine blades, characterized in that, It includes an electrode network, a current collection module, a pre-processing module, an identification module and an alarm module; the electrode network includes a plurality of micro-electrodes printed or embedded in the surface of the turbine blade to be tested with a preset density distribution; the current collection module is connected to the electrode network through a brush and a conductive structure; The current collection module is used to receive current data of the electrode network when the turbine blade to be tested rotates; And, obtaining the resistance distribution information of the turbine blade to be tested based on the current data and the electrode distribution information of the electrode network; Sending the resistance distribution information to the preprocessing module; The preprocessing module is used to preprocess the resistance distribution information and then send it to the identification module; The recognition module is used to input the preprocessed resistance distribution information into a defect recognition model to obtain a recognition result; The alarm module is used to generate alarm prompt information when the recognition result meets the preset alarm type.
2. The real-time non-destructive testing system for turbine blades according to claim 1, wherein Also includes training modules; The training module is used to obtain normal resistance distribution information of normal turbine blades under different working conditions and defect resistance distribution information of defective turbine blades with different defect types; and send the information to the preprocessing module; and obtaining defect parameter information of each defective turbine blade; Generate corresponding annotation data of defect resistance distribution information based on the defect parameter information; Acquire a pre-trained model, and train the pre-trained model based on the labeled data, the received pre-processed normal resistance distribution information, and the defect resistance distribution information to obtain the defect recognition model; The preprocessing module is further used to preprocess the received normal resistance distribution information and the defective resistance distribution information and then send them to the training module.
3. The real-time non-destructive testing system for turbine blades according to claim 1, wherein, The current collecting module and the brush on the rotor where the turbine blade to be tested is located are connected by sliding contact; the conductive structure extends along the surface of the rotor to the end of the rotor and is connected to the brush at the end of the rotor.
4. The real-time non-destructive testing system for turbine blades according to claim 1, wherein, The preprocessing module is specifically used to treat the resistance distribution information exceeding a preset threshold range as an abnormal value and eliminate it.
5. The real-time non-destructive testing system for turbine blades according to claim 1, wherein The preprocessing module is specifically used to determine the missing data points in the resistance distribution information; generate interpolation data according to the adjacent data of the missing data points and using a linear interpolation method or a mean filling method, and use the interpolation data as the missing data points.
6. The real-time non-destructive testing system for turbine blades according to claim 1, characterized in that, The preprocessing module is specifically used to normalize the resistance distribution information.
7. The real-time non-destructive testing system for turbine blades according to claim 2, characterized in that, The defect parameter information includes the crack length, crack depth, corrosion area and wear amount of the defective turbine blade.
8. The real-time non-destructive testing system for turbine blades according to claim 7, wherein The annotated data includes blade geometry, defect type, severity, manufacturing method and acquisition time.
9. The real-time non-destructive testing system for turbine blades according to claim 2, characterized in that, The pre-trained model is a deep neural network model, and the activation function is a ReLU function.
10. The real-time non-destructive testing system for turbine blades according to claim 1, characterized in that, The alarm module is specifically used to output the alarm prompt information in the form of sound and light alarm, interface pop-up window or SMS push.