Ultrasonic water immersion flaw detection method using intelligent sensor

Abstract

The invention discloses an ultrasonic water immersion flaw detection method applying an intelligent sensor, and relates to the technical field of intelligent sensors. The detection method comprises the following steps: according to specification information of a to-be-detected target workpiece, screening out a corresponding intelligent sensor and a working frequency thereof from a pre-constructed matching sensing library; according to the technical key points, the CNN is introduced into a complex scene where ERF is prone to misjudgment, the correction capacity of key misjudgment points is remarkably improved, the CNN is locally called only when necessary, high computing power consumption of the whole area is avoided, the overall efficiency is still close to that of a pure ERF scheme, through confidence feedback, the method can be automatically focused on a fuzzy area, the detection rate of tiny or complex defects is improved, and the method is suitable for large-scale popularization and application. The operation of the whole scheme realizes dynamic collaborative operation, while certain accuracy is ensured, the detection efficiency and pertinence can be ensured, the recognition degree of defect types is improved, and the balance stability among the three effects is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent sensors, and in particular to an ultrasonic water immersion flaw detection method using intelligent sensors. Background Art

[0002] When used in ultrasonic water immersion testing, smart sensors typically refer to sensor units that integrate data acquisition, signal processing, and communication functions, with capabilities such as self-calibration and environmental compensation. The current conventional testing process involves immersing the workpiece in a water tank, transmitting ultrasonic waves through water coupling. A mechanical device drives the probe to scan the workpiece surface along a predetermined path. Ultrasonic pulses are emitted, and reflected echoes are received. Echo characteristics such as echo time and amplitude are analyzed to determine the location and size of the defect. An imaging system is then used to generate C-scan or B-scan images, and the nature of the defect is determined manually or by software. The entire process relies on fixed parameter settings, with the scanning path and resolution typically pre-set and lacking real-time adjustment capabilities.

[0003] The following problems may occur when using traditional existing or designed flaw detection methods: Traditional ultrasonic water immersion flaw detection mostly adopts a fixed scanning strategy, using a uniform resolution regardless of regional characteristics, which leads to frequent waste of resources and missed detections. For example, when inspecting thick-walled workpieces, high-resolution scans are still performed on areas without obvious abnormalities, which is time-consuming and consumes a lot of computing resources. However, in complex structural areas such as weld corners, the insufficient resolution makes it difficult to identify tiny cracks, resulting in misjudgments or missed detections. Although there are currently some methods that introduce automatic recognition algorithms on the market, most of them rely on a single model for full operation and cannot be dynamically adjusted according to actual needs, resulting in a lack of effective review mechanisms in relevant areas prone to misjudgment. In addition, traditional methods lack the ability to process regional differentiation and cannot achieve flexible scheduling, making it impossible or difficult to effectively balance efficiency and accuracy. If manual review is used, it will not only increase the manual review burden but also affect the reliability of the overall detection. This is especially urgent in large-scale or fast-paced production scenarios, where these issues are even more urgent. Summary of the Invention

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: An ultrasonic water immersion flaw detection method using an intelligent sensor includes the following steps: Based on the specifications of the target workpiece to be inspected, the corresponding smart sensor and its operating frequency are selected from the pre-built matching sensor library. A dynamic area partitioning strategy is used to obtain a partitioning scheme for the target workpiece to be inspected. Each scanning area is divided, and a probability model is established to output the probability of the calibrated defect type and its possible location. The output of the probability model is positively correlated with the priority, and a high- or low-resolution scanning strategy is executed based on the priority. Introducing a real-time data analysis mechanism to perform preliminary defect identification actions on scanning areas of different priorities; When the low-resolution scanning strategy identifies a suspected defect, it triggers a strategy change action; Scanning areas that implement high-resolution scanning strategies are marked as critical areas, while those that implement low-resolution scanning strategies are marked as non-critical areas. A dual-model collaborative architecture is adopted: a primary model and a secondary model. A scheduling management mechanism is introduced as a constraint. When using the secondary model, the constraints are applied to complete the adjustment action and generate the final probability. When the final probability exceeds the judgment threshold, it is marked as a defect; otherwise, no mark is made.

[0005] Furthermore, in the matching sensor library: when detecting a target workpiece with a maximum thickness exceeding 5mm, the operating frequency of the benchmark is: 1MHz; when detecting a target workpiece with a maximum thickness not exceeding 5mm, the operating frequency of the benchmark is: 5MHz; The target workpiece thickness is inversely proportional to the value of the operating frequency setting.

[0006] Furthermore, the dynamic area division strategy adopted is as follows: the geometric information of the target workpiece surface is obtained through three-dimensional scanning, and the partitioning scheme is generated using computer-aided design (CAD) software; the process of executing high or low-resolution scanning strategies according to priority is as follows: when the priority exceeds the calibration value, it is marked as high priority and the high-resolution scanning strategy is executed; otherwise, it is marked as low priority and the low-resolution scanning strategy is executed.

[0007] Furthermore, the main model uses a lightweight convolutional neural network, namely CNN, to deal with key areas; the auxiliary model uses enhanced random forest, namely ERF, to deal with non-critical areas.

[0008] Furthermore, the operation process of the scheduling management mechanism: A1: data complexity assessment; A2: confidence and uncertainty processing; A3: dynamic decision-making mechanism for CNN participation ratio.

[0009] Furthermore, the data complexity assessment process mentioned in A1 is: analyze the signal corresponding to the non-critical area, obtain the complexity assessment index, including at least: noise level, frequency component change and energy distribution, perform weighted calculation based on the complexity assessment index, and obtain the complexity value; when the complexity value exceeds the preset first standard threshold, the first linear proportional adjustment strategy is triggered; otherwise, no response action is taken.

[0010] Furthermore, noise level: the peak signal-to-noise value of the signal is calculated as the quantitative value of the noise level; frequency component change: the fast Fourier transform (FFT) is used to analyze the spectral distribution of the signal, and for continuous segments in the signal in non-critical areas, the difference in the spectral center frequency between adjacent segments is calculated, and the maximum value is screened as the quantitative value of the frequency component change; energy distribution: for signals in non-critical areas, its total energy is calculated, and the information entropy is obtained based on the total energy. The inverse of the information entropy is used to reflect the concentration of the energy distribution and is used as the quantitative value of the energy distribution.

[0011] Furthermore, the triggered first linear proportional adjustment strategy process is: based on the complexity value C, the proportion of the introduction of lightweight convolutional neural network participation is obtained according to the method: β_C=min(1, max(0, k_C×(C-C_thd))); where β_C is the CNN proportion based on the complexity value, k_C is the adjustment coefficient, the value range is [0, 2], and C_thd is the first standard threshold.

[0012] Furthermore, the confidence and uncertainty processing process mentioned in A2 is: analyze the confidence of the corresponding auxiliary model output, and compare the confidence with the preset second standard threshold. When the confidence does not exceed the second standard threshold, the second linear proportional adjustment strategy is triggered; otherwise, no response action is taken; wherein, the triggered second linear proportional adjustment strategy process is: based on the confidence D, the proportion of the introduction of lightweight convolutional neural network participation is obtained, and the method is based on: β_D=max(0, 1-D / D_thd); where β_D is the CNN ratio based on confidence, and D_thd is the second standard threshold.

[0013] Furthermore, the dynamic decision-making mechanism for the CNN participation ratio mentioned in A3 is as follows: when any adjustment strategy is triggered, the corresponding adjustment strategy result is used as the basis for the fusion probability model, and the fusion probability model is run to output the final probability; when the two first and second linear proportional adjustment strategies are triggered at the same time, the maximum value is taken as the basis for the fusion probability model, and the fusion probability model is run: P_final=a×P_CNN+(1-a)×P_ERF, and the final probability P_final is output; where the value of a is β_C or β_D.

[0014] The present invention provides an ultrasonic water immersion flaw detection method using an intelligent sensor, which has the following beneficial effects: 1) This solution combines intelligent sensors with dynamic area division, using different scanning resolutions for different areas during the inspection process. This differentiated processing not only improves the ability to identify subtle defects and effectively reduces false alarms, but also saves operating resources. This not only ensures the accuracy of inspection results, but also ensures the efficiency and reliability of the overall inspection process. 2) This solution utilizes high- and low-resolution scanning strategies and introduces a real-time data analysis mechanism. Initial detection of suspected defects triggers a higher-resolution review, while dynamically adjusting subsequent scanning plans based on actual needs. This not only improves the overall inspection solution's runtime response speed and flexibility, but also reduces computing costs to a certain extent, enabling rational resource planning and utilization, and resolving the irrational resource allocation problem caused by fixed resource allocation in traditional methods. 3) This solution adopts a dual-model collaborative architecture, combined with an intelligent scheduling algorithm under constraints. When specific conditions are met, the collaboration between the two models is dynamically adjusted to cope with complex and changing actual working conditions. This not only enhances the robustness and adaptability of the overall solution, but also provides users with a flexible and efficient detection method. At the same time, the dual-model collaborative architecture, dynamic area division, and enhanced recognition solutions form an overall closed loop, realizing intelligent management of the entire process from data collection to final decision-making, ensuring the efficiency and flexibility of the ultrasonic water immersion flaw detection process. 4) This solution introduces CNN in complex scenarios where ERF is prone to misjudgment, significantly improving the ability to correct key misjudgment points. CNN is only called locally when necessary to avoid high computing power consumption in the entire area. The overall efficiency remains close to that of a pure ERF solution. Through confidence feedback, it can automatically focus on fuzzy areas, improving the detection rate of small or complex defects. The operation of the overall solution realizes dynamic collaborative operation, while ensuring a certain level of accuracy, it can also maintain detection efficiency and targeted improvement of defect type recognition, ensuring a balanced and stable balance between the three effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Schematic diagram of the brief steps of the detection method of the present invention. DETAILED DESCRIPTION

[0016] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0017] See also Figure 1 This embodiment provides an ultrasonic water immersion flaw detection method using an intelligent sensor. This detection method is mostly used for flaw detection of steam turbine blades. The specific steps of the detection method are described as follows: S1. Initialization configuration: Based on the specifications of the target workpiece to be inspected, the corresponding smart sensors and their operating frequencies are selected from the pre-built matching sensor library. The type of target workpiece to be inspected can be a turbine blade or an engine or gas turbine component, etc., and the type can be selected according to actual needs. The smart sensors used in this embodiment are mostly cost-effective mid-range sensors. To deal with these types of target workpieces, it is necessary to ensure basic detection accuracy while reducing costs. For example, in this embodiment, in the ultrasonic water immersion flaw detection scenario, a sensor made of piezoelectric ceramic material with high sensitivity and wide-band response is selected as the smart sensor to provide clear and detailed signal feedback, which is particularly important for detecting subtle defects. In addition, sensor systems that support modular design can also be selected as needed to facilitate flexible configuration adjustment according to different detection needs. This will not be described in detail here. In the matching sensor library, set: When detecting a target workpiece with a maximum thickness exceeding 5mm, the operating frequency of the benchmark is 1MHz, which can provide better penetration capability. When detecting a target workpiece with a maximum thickness not exceeding 5mm, the operating frequency of the benchmark is 5MHz, and a higher frequency is required to achieve higher lateral resolution, which is more suitable for detecting subtle defects. The thicker the target workpiece, the smaller the operating frequency setting, and the two are inversely proportional. In actual settings, in order to verify this theory, a series of experiments can be conducted to compare the detection effects at different frequencies, so as to ultimately determine the optimal frequency that best suits the current application scenario. Of course, these series of experiments are entered into the matching sensor library after confirmation.

[0018] S2. Data segmentation analysis: S2.1. Dynamic area division: In traditional methods, a uniform grid or a fixed-size rectangular frame is usually used to divide the area to be inspected. However, when facing workpieces with irregular shapes or large thickness variations, this method may cause some areas to be over-scanned while other important areas are ignored. Therefore, in the specific implementation scheme of S2 in this embodiment, the following is proposed: Dynamic area partitioning strategy: The content of this partitioning strategy is: obtain the geometric information of the target workpiece surface through 3D scanning, and use computer-aided design (CAD) software to generate the optimal scanning path and partitioning scheme; For example, for aircraft engine blades, the root and tip parts may require higher resolution for inspection because these areas are more prone to cracks caused by stress concentration. In contrast, the middle part can be scanned quickly using a lower resolution. This differentiated processing can significantly improve overall efficiency without affecting the final inspection results. S2.2. Prioritization of defects under prior guidance: Using historical data and expert knowledge, a probability model is established for common defect types (i.e., calibration defect types) and their likely locations. Prioritize each scan area based on the output of this probability model. The output of the probability model is positively correlated with the priority. The higher the probability, the more severe the potential risk, and the higher the corresponding priority. When the priority exceeds the calibration value, it is marked as high priority and the high-resolution scanning strategy is executed; When the priority does not exceed the calibration value, it is marked as low priority and a low-resolution scanning strategy is executed; A high-resolution scanning strategy is implemented in high-priority corresponding areas, while low-resolution scanning is used in high-priority corresponding areas. This overall operating strategy enables the solution to focus more on the detection of potentially serious defects, improves the recognition of specific types of defects, and reduces excessive attention to non-critical areas, achieving the purpose of accurate positioning and assessment.

[0019] For example, on aircraft engine blades, fatigue cracks often appear near stress concentration points; while at weld joints, problems such as porosity or lack of fusion are more likely to occur. Based on such a probabilistic model, we can sort each scanning area according to its potential risk level and prioritize those areas most likely to have serious defects. For example, if it is known that a certain model of engine blade has failed multiple times in the past few years due to cracks in a specific location, then during regular maintenance inspections of this model of blade, this location will automatically be set as the highest priority. Even if it is located in a difficult-to-reach place, a special high-precision scanning task will be arranged. Among them, the resolution exceeding the calibrated resolution index is high resolution, and vice versa, it is low resolution. There are clear definitions and divisions for this, and we will not elaborate on it or give examples here. Effect description: By applying intelligent sensors and dynamic area division, the present invention can adopt different scanning resolutions for different areas during the detection process. This differentiated processing not only improves the ability to identify subtle defects and effectively reduces the false alarm rate, but also saves certain operating resources. On the one hand, it ensures the accuracy of the detection results, and on the other hand, it ensures the efficiency and reliability of the overall detection process.

[0020] S2.3, Real-time feedback and adaptive adjustment: In actual operation, a real-time data analysis mechanism is introduced to perform preliminary defect identification actions on scanning areas of different priorities. When the low-resolution scanning strategy identifies a suspected defect, a strategy change action is triggered, changing the original low-resolution scanning strategy to a high-resolution scanning strategy. The specific display method is: allowing the system to dynamically adjust the subsequent scanning plan according to the current situation; For example, if a suspected major defect is unexpectedly discovered in a low-priority area, the original plan is immediately suspended and the importance of the area is reassessed. In this embodiment, the reassessment is not performed, but the scanning strategy is directly changed because this is more efficient. Under other more necessary conditions, more resources are added to conduct a deeper inspection. For example, during a routine inspection, although the initial plan was to conduct a comprehensive but relatively rough scan of the entire blade, some abnormal signals were detected when scanning a small area near the edge. The system then suspended scanning of other areas and conducted a higher-resolution review of this area, ultimately confirming a small, hidden crack and avoiding a potential safety hazard. Expansion plan description: After S2.3, operations related to resource management and optimized scheduling can be added; The specific operation process is as follows: According to the configuration of the configured smart sensor, including the maximum sampling rate and processing speed, a resource allocation strategy is formulated: when the current processing speed is lower than the target setting value, the spare smart sensor is rented through the cloud computing platform for processing, and the number of rented sensors is increased according to the gradient setting amount until the current processing speed is not lower than the target setting value; when faced with an emergency, the detection accuracy of some low-priority scanning areas is temporarily sacrificed to ensure that the high-priority scanning areas are fully inspected; the specific instructions are: during peak hours, additional computing power can be rented through the cloud computing platform to speed up the data analysis process; or when faced with an emergency, the detection accuracy of some non-critical areas can be temporarily sacrificed to ensure that the core parts are fully inspected. Inspection; For example: In a large manufacturing plant, due to a backlog of orders, the quality inspection of a batch of newly produced engine blades must be completed as soon as possible; at this time, the factory can choose to hand over a portion of simpler and lower-risk blades to an external cloud service provider for rapid initial screening, retaining only the most challenging samples for detailed analysis in the local laboratory, thereby maximizing efficiency; by adopting the above-mentioned expansion solution to achieve improvements, not only can work efficiency be greatly improved while maintaining high detection accuracy, but also more accurate positioning and evaluation of specific types of defects can be provided. This flexible and changeable strategy enables calm response to various complex situations in both routine maintenance and emergency repair scenarios, reflecting the demand for intelligent and personalized services.

[0021] Effect description: In combination with the high- and low-resolution scanning strategies mentioned above, a real-time data analysis mechanism is further introduced to trigger a higher-resolution review when a suspected defect is initially detected. At the same time, the subsequent scanning plan is dynamically adjusted according to actual needs. This can not only improve the response speed and flexibility of the overall inspection solution during operation, but also reduce computing costs to a certain extent, realize the rational planning or utilization of resources, and solve the problem of unreasonable allocation caused by fixed resource division in traditional methods.

[0022] S3. Defect classification and identification: S3.1. Construction of dual-model collaborative architecture: The scanning area that ultimately executes the high-resolution scanning strategy is marked as the critical area, and the scanning area that ultimately executes the low-resolution scanning strategy is marked as the non-critical area. A dual-model collaborative architecture is adopted: a main model and a secondary model; Among them, the conventional solution is: the main model can use lightweight convolutional neural network, namely Lightweight CNN, to deal with key areas; the auxiliary model can use enhanced random forest, namely ERF, to deal with non-critical areas; It should be noted that a lightweight convolutional neural network is used to process critical areas. Its model architecture uses depthwise separable convolution, which significantly reduces the number of parameters and computational complexity while retaining sensitivity to spatial features such as crack direction and delamination boundaries. The input is ultrasonic A-scan and B-scan signals (i.e., the original signals), and the output is the probability of defect existence and preliminary classification. Preliminary categories include cracks, pores, inclusions, and delamination. The lightweight convolutional neural network can achieve millisecond-level inference in a GPU-accelerated environment, making it fully suitable for such critical areas with high accuracy requirements. The enhanced random forest is used to process non-critical areas. Its input is an engineered feature vector extracted from the original signal, which includes signal peak, rise time, energy integral, spectral center of gravity, and wavelet coefficient entropy. The enhanced random forest performs well when the feature dimension is low, and the training and inference speed are fast. It is suitable for deployment on edge computing devices. It has strong recognition capabilities for specific defect patterns, such as the high-frequency attenuation characteristics of welding pores, and is robust to noise.

[0023] S3.2, Dispatching Management Operation: A scheduling management mechanism is introduced as a constraint. When using the auxiliary model, the constraint is applied to complete the adjustment action. The operation process of the scheduling management mechanism is as follows: S3.2.1. Data complexity assessment: Analyze the signals corresponding to the non-critical areas to obtain complexity evaluation indicators, including noise level, frequency component changes, and energy distribution. Each evaluation indicator has a corresponding quantitative value. A weighted calculation is performed based on the complexity evaluation indicators to obtain a comprehensive score as the complexity value. If the complexity value exceeds a preset first standard threshold, a first linear proportional adjustment strategy is triggered; otherwise, no response action is taken. The above mentioned signals are defined as follows: The original signals obtained from the ultrasonic water immersion flaw detection process, namely the ultrasonic A-scan and B-scan signals, also contain information about the internal structure of the target workpiece, including the time, amplitude, and frequency components of the reflected echo. The corresponding non-critical area signal is the detection result for non-critical areas; Noise level: measured by calculating the peak signal-to-noise ratio of the signal, which is used as a quantitative value of the noise level; where the peak signal-to-noise ratio is defined as: PSNR = 10 × log 10 (Ma I 2 / MSE); where PSNR is the peak signal-to-noise ratio, Ma I is the maximum possible value of the signal, and MSE is the mean square error between the signal in the non-critical area and the smooth background. Therefore, the lower the PSNR, the higher the noise. Frequency component change: Use the fast Fourier transform (FFT) to analyze the spectral distribution of the signal. For continuous segments in the signal of the non-critical area, calculate the difference in the center frequency of the spectrum between adjacent segments, and select the maximum value as the quantized value of the frequency component change. Among them, the fast Fourier transform (FFT) is used to convert the time domain signal into the frequency domain signal, and the center frequency of the spectrum, that is, the center of gravity of the spectrum, is calculated, which is defined as: ; Where CF represents the center of gravity of the spectrum, f i is the i-th frequency component, X(f i ) is the complex amplitude at the i-th frequency. Calculate the difference in spectral center frequencies between adjacent segments, that is, the difference between the spectral centers of adjacent frequency bands. Ensure that the difference is not negative by calculating the absolute value. Filter the maximum value among the differences as the quantized value of the frequency component change, expressed as the frequency component change FC. Energy distribution: For signals in non-critical areas, calculate their total energy and obtain information entropy based on the total energy. Use the inverse of the information entropy to reflect the concentration of energy distribution and use it as the quantized value of energy distribution. The formula for calculating the total energy is: Where E is the total energy, x[n] is the time series sample value of the signal; the process of obtaining information entropy based on the total energy is as follows: the signal corresponding to the non-critical area is divided into multiple sub-intervals, and the energy of each interval is calculated separately. j, calculate the normalized energy distribution p j =E j / E, and then calculate the information entropy Ey: Ey=-∑p j log(p j ); where j represents the number of each subinterval. A higher information entropy Ey indicates a more uniform energy distribution, while a lower information entropy Ey indicates that the energy is concentrated in certain specific areas. Therefore, the final quantified value of energy distribution is -Ey, which can effectively reflect the concentration of energy distribution. The sum of the weights used in weighted calculations based on complexity evaluation indicators is 1. In practical applications, it should be noted that all complexity evaluation indicators should be normalized to the same range, for example, [0, 1], so that different complexity evaluation indicators can be directly added together. If the original range of a complexity evaluation indicator is large, it can be scaled to an appropriate range through linear transformation or other normalization methods. It should be noted that the preset first standard threshold is a critical indicator for determining whether the first linear proportional adjustment strategy is triggered. The setting of its value is based on historical data analysis. By calculating the complexity value C of known samples and combining it with the detection results, a value is selected that can distinguish between simple and complex data. For example, 0.6 can be selected so that when the complexity value C exceeds this value, the corresponding strategy is triggered. The purpose is to ensure that defects can still be effectively detected under complex backgrounds. The first standard threshold needs to be adjusted and optimized according to the specific application scenarios and workpiece characteristics. The principles of the second standard threshold and the judgment threshold mentioned below are similar or the same, so they will not be elaborated here.

[0024] Effect description: The present invention adopts a dual-model collaborative architecture and combines it with an intelligent scheduling algorithm under constraint conditions; when specific conditions are met, the collaboration mode between the two models will be dynamically adjusted to cope with complex and changeable actual working conditions. On the one hand, it can enhance the robustness and adaptability of the overall scheme operation, and on the other hand, it also provides users with a flexible and efficient detection method. At the same time, the above-mentioned dual-model collaborative architecture and the dynamic area division, enhanced recognition and other schemes mentioned in the above steps form an overall closed-loop system, realizing intelligent management of the entire process from data acquisition to final decision-making, ensuring the efficiency and flexibility of the ultrasonic water immersion flaw detection process.

[0025] The first linear scaling strategy is triggered as follows: The proportion of lightweight convolutional neural network participation is obtained based on the complexity value C, and the method is: β_C=min(1, max(0, k_C×(C-C_thd))); where β_C is the CNN ratio based on the complexity value, k_C is the adjustment coefficient, which controls the response slope and has a value range of [0, 2], which is 2 in this embodiment; C_thd is the first standard threshold.

[0026] S3.2.2. Confidence and uncertainty handling: Analyze the confidence level of the auxiliary model output and compare it with the preset second standard threshold. If the confidence level does not exceed the second standard threshold, it is considered that there is recognition uncertainty in the corresponding non-critical area, and the second linear proportional adjustment strategy is triggered. Otherwise, no response action is taken. The second linear proportional adjustment strategy is triggered as follows: The proportion of lightweight convolutional neural networks introduced based on the confidence D is obtained according to the following method: β_D=max(0,1-D / D_thd); where β_D is the CNN ratio based on the confidence, and D_thd is the second standard threshold; when the confidence D approaches 0, β_D tends to 1; when the confidence D does not exceed the second standard threshold, β_D=0; S3.2.3 Dynamic decision-making mechanism for CNN participation ratio: When only any one of the above adjustment strategies is triggered, the corresponding adjustment strategy result is used as the basis for the fusion probability model, and the fusion probability model is run to output the final probability. When both of the above adjustment strategies are triggered at the same time, the maximum value is taken as the basis for the fusion probability model, and the fusion probability model is run: P_final=a×P_CNN+(1-a)×P_ERF, and the final probability P_final is output. When the final probability exceeds the judgment threshold, it is marked as a defect. When the final probability does not exceed the judgment threshold, no mark is made. The value of a in the above formula can be: β_C or β_D. Among them, if any of the adjustment strategies triggered is: the first linear proportional adjustment strategy, the corresponding adjustment strategy result is: the CNN proportion β_C based on the complexity value; when running the fusion probability model, the method is based on: P_final=β_C×P_CNN+(1-β_C)×P_ERF; if any of the adjustment strategies triggered is: the second linear proportional adjustment strategy, the corresponding adjustment strategy result is: the CNN proportion β_D based on the confidence value; when running the fusion probability model, the method is based on: P_final=β_D×P_CNN+(1-β_D)×P_ERF; when the first and second linear proportional adjustment strategies are triggered at the same time, the maximum value is taken as the basis of the fusion probability model, and the specific method is: a=max(β_C, β_D).

[0027] Effect description: The present invention introduces a scheduling management mechanism, adopts a dual-model collaborative architecture and a dynamic adjustment strategy. On the one hand, it triggers lightweight CNN fusion analysis based on complexity evaluation and confidence judgment in non-critical areas, thereby improving the recognition accuracy of potential defects. On the other hand, it avoids high computing power overhead in the entire area and significantly improves the detection efficiency. The application of the overall solution solves the problem that traditional ultrasonic flaw detection is difficult to strike a balance between efficiency and accuracy, especially when the background is complex or the signal is fuzzy. It can still maintain high robustness, and through weighted complexity scoring and linear proportional adjustment, it realizes refined control of model collaboration, reflecting the adaptive ability of the intelligent detection solution in resource allocation and decision optimization, forming a closed-loop management from perception to decision-making, and ensuring the efficiency and reliability of the detection process.

[0028] It should be noted that the policy execution logic for different regions is: For critical areas, CNN is used: high-precision full-process processing is maintained, and the above-mentioned S3.2 and scheduling management mechanisms are not introduced. For non-critical areas, ERF is used: under normal circumstances, only the ERF model is run. Only when the data complexity assessment or confidence constraint (corresponding to S3.2.1 and S3.2.2) is triggered, CNN is automatically called to perform additional analysis on the current local area, and the fusion ratio is calculated according to the above formula to generate the final judgment. Example: In the initial ERF analysis of a non-critical area of ​​a blade or workpiece, a point is detected with a confidence of only 0.55, which does not exceed the second standard threshold of 0.7. The complexity value is 0.72, which exceeds the preset first standard threshold of 0.6. In this case, the fusion probability model needs to be run: Calculation: β_C = 2×(0.72-0.6) = 0.24; β_D = 1-0.55 / 0.7≈0.214 (retain two decimal places), and the final a = max(0.24, 0.21) = 0.24; the output probability after CNN analysis is: 0.92, and the ERF output is: 0.65. The final probability P_final: P_final = 0.24×0.92+(1-0.24)×0.65≈0.71 (retain two decimal places). Since 0.71 exceeds the judgment threshold of 0.7, it is marked as a defect.

[0029] By adopting the above technical solutions, a balance is achieved among accuracy assurance, efficiency control and specific recognition enhancement. Among them, accuracy assurance: CNN is introduced in complex scenarios where ERF is prone to misjudgment, significantly improving the ability to correct key misjudgment points; efficiency control: CNN is only called locally when necessary to avoid high computing power consumption in the entire area, and the overall efficiency is still close to that of a pure ERF solution; specific recognition enhancement: through confidence feedback, it can automatically focus on fuzzy areas to improve the detection rate of small or complex defects. The operation of the overall solution realizes dynamic collaborative operation. For flaw detection operations, it does not mean that flaw detection needs to ensure the highest accuracy. In this solution, while ensuring a certain accuracy, it can also ensure detection efficiency and targeted improvement of defect type recognition, ensuring a stable balance between the three effects.

[0030] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0031] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0032] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. An ultrasonic water immersion flaw detection method using an intelligent sensor, the method comprising the following steps: According to the specification information of the target workpiece to be detected, the corresponding smart sensor and its operating frequency are screened from a pre-built matching sensor library; the detection method further comprises the following steps: A dynamic area division strategy is used to obtain a partitioning scheme for the target workpiece to be inspected: each scanning area is divided, a probability model is established, and the probability of the calibrated defect type and its possible location is output. The output of the probability model is positively correlated with the priority, and a high or low resolution scanning strategy is executed according to the priority. Introducing a real-time data analysis mechanism to perform preliminary defect identification actions on scanning areas of different priorities; When the low-resolution scanning strategy identifies a suspected defect, it triggers a strategy change action; Scanning areas that implement high-resolution scanning strategies are marked as critical areas, while those that implement low-resolution scanning strategies are marked as non-critical areas. A dual-model collaborative architecture is adopted: a primary model and a secondary model. A scheduling management mechanism is introduced as a constraint. When using the secondary model, the constraints are applied to complete the adjustment action and generate the final probability. When the final probability exceeds the judgment threshold, it is marked as a defect; otherwise, no mark is made.

2. The ultrasonic water immersion flaw detection method using an intelligent sensor according to claim 1, characterized in that: In the matching sensor library: when detecting a target workpiece with a maximum thickness exceeding 5mm, the operating frequency of the calibration is: 1MHz; when detecting a target workpiece with a maximum thickness not exceeding 5mm, the operating frequency of the calibration is: 5MHz; The target workpiece thickness is inversely proportional to the value of the operating frequency setting.

3. The ultrasonic water immersion flaw detection method using an intelligent sensor according to claim 1, characterized in that: The dynamic area partitioning strategy adopted is as follows: the geometric information of the target workpiece surface is obtained through three-dimensional scanning, and the partitioning scheme is generated using computer-aided design (CAD) software; the process of executing high or low resolution scanning strategies according to priority is as follows: when the priority exceeds the calibration value, it is marked as high priority and the high-resolution scanning strategy is executed; otherwise, it is marked as low priority and the low-resolution scanning strategy is executed.

4. The ultrasonic water immersion flaw detection method using an intelligent sensor according to claim 1, characterized in that: The main model uses a lightweight convolutional neural network, or CNN, to deal with key areas; the auxiliary model uses enhanced random forest, or ERF, to deal with non-critical areas.

5. The ultrasonic water immersion flaw detection method using an intelligent sensor according to claim 4, characterized in that: The operation process of the scheduling management mechanism: A1: Data complexity assessment; A2: Confidence and uncertainty processing; A3: Dynamic decision-making mechanism for CNN participation ratio.

6. The ultrasonic water immersion flaw detection method using an intelligent sensor according to claim 5, characterized in that: The data complexity assessment process mentioned in A1 is: analyze the signal corresponding to the non-critical area, obtain the complexity assessment index, including at least: noise level, frequency component change and energy distribution, perform weighted calculation based on the complexity assessment index, and obtain the complexity value; when the complexity value exceeds the preset first standard threshold, the first linear proportional adjustment strategy is triggered; otherwise, no response action is taken.

7. The ultrasonic water immersion flaw detection method using an intelligent sensor according to claim 6, characterized in that: Noise level: The peak signal-to-noise value of the signal is calculated as the quantitative value of the noise level; Frequency component change: The fast Fourier transform (FFT) is used to analyze the spectral distribution of the signal. For continuous segments in the signal in non-critical areas, the difference in the spectral center frequency between adjacent segments is calculated, and the maximum value is selected as the quantitative value of the frequency component change; Energy distribution: For signals in non-critical areas, its total energy is calculated, and the information entropy is obtained based on the total energy. The inverse of the information entropy is used to reflect the concentration of the energy distribution and is used as the quantitative value of the energy distribution.

8. The ultrasonic water immersion flaw detection method using an intelligent sensor according to claim 5, characterized in that: The triggered first linear proportional adjustment strategy process is: based on the complexity value C, the proportion of the introduction of lightweight convolutional neural network participation is obtained according to the method: β_C=min(1, max(0, k_C×(C-C_thd))); where β_C is the CNN ratio based on the complexity value, k_C is the adjustment coefficient, the value range is [0, 2], and C_thd is the first standard threshold.

9. The ultrasonic water immersion flaw detection method using an intelligent sensor according to claim 8, characterized in that: The confidence and uncertainty processing process mentioned in A2 is: analyze the confidence of the corresponding auxiliary model output, and compare the confidence with the preset second standard threshold. When the confidence does not exceed the second standard threshold, the second linear proportional adjustment strategy is triggered; otherwise, no response action is taken; wherein, the triggered second linear proportional adjustment strategy process is: based on the confidence D, the proportion of the introduction of lightweight convolutional neural network participation is obtained, and the method is based on: β_D=max(0, 1-D / D_thd); where β_D is the CNN ratio based on confidence, and D_thd is the second standard threshold.

10. The ultrasonic water immersion flaw detection method using an intelligent sensor according to claim 9, characterized in that: The dynamic decision-making mechanism for the CNN participation ratio mentioned in A3 is as follows: when any adjustment strategy is triggered, the corresponding adjustment strategy result is used as the basis for the fusion probability model, and the fusion probability model is run to output the final probability; when the first and second linear proportional adjustment strategies are triggered at the same time, the maximum value is taken as the basis for the fusion probability model, and the fusion probability model is run: P_final=a×P_CNN+(1-a)×P_ERF, and the final probability P_final is output; where the value of a is β_C or β_D.

Images