A machine learning based phased array ultrasound parameter adjustment method and system

Abstract

The present application relates to the technical field of parameter adjustment, in particular to a phased array ultrasonic parameter adjustment method and system based on machine learning; the method comprises obtaining ultrasonic echo signals of a weld in each detection period within a period of time and phased array ultrasonic parameters in a machine learning processing platform; based on each ultrasonic echo signal, determining each echo characteristic information reflecting echo quality; based on each echo characteristic information, determining a change state of echo quality change; based on the change state of echo quality change, determining a parameter adjustment scheme; using the parameter adjustment scheme to adjust the phased array ultrasonic parameters to obtain adjusted phased array ultrasonic parameters, so as to complete the adjustment of the phased array ultrasonic parameters. The present application aims to solve the problem that the existing parameter adjustment method is difficult to accurately adjust the phased array ultrasonic parameters, resulting in difficulty in accurately detecting the tiny defects of the weld and low detection accuracy.

Description

Technical Field

[0001] This invention relates to the field of parameter adjustment technology, and specifically to a method and system for adjusting phased array ultrasonic parameters based on machine learning. Background Technology

[0002] When performing non-destructive testing on welds, the phased array ultrasonic equipment is used to scan and inspect the workpiece. Based on the real-time acquired echo signals, quality evaluation indicators such as signal-to-noise ratio and defect location consistency are extracted, and then the transmission angle, center frequency, and receiving gain of the phased array ultrasonic equipment are adjusted.

[0003] However, when scanning and inspecting thick-walled welds, the probe of the phased array ultrasonic equipment needs to pass through the coarse-grained heat-affected zone. The coarse-grained material causes strong background noise, leading to random fluctuations in the signal-to-noise ratio. Simultaneously, the scattering of the sound beam by the coarse grains causes non-systematic fluctuations in the amplitude of defect echoes and structural echoes. Existing parameter adjustment methods struggle to precisely adjust the phased array ultrasonic parameters. The received quality evaluation score no longer cleanly reflects the adjustment effect of a single parameter, but rather is a mixture of the combined effects of the emission angle, frequency, gain, and material noise disturbances. When the quality evaluation score deteriorates, it becomes impossible to accurately determine whether it is due to angle deviation, frequency detuning, or insufficient gain, resulting in directional ambiguity in the internal gradient estimation. This makes accurate detection of minute defects in the weld difficult, leading to low detection accuracy. Summary of the Invention

[0004] The purpose of this invention is to provide a phased array ultrasonic parameter adjustment method and system based on machine learning, which solves the problem that existing parameter adjustment methods are difficult to accurately adjust phased array ultrasonic parameters, resulting in difficulty in accurately detecting minute defects in welds and low detection accuracy.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for adjusting phased array ultrasonic parameters based on machine learning, comprising: Acquire the ultrasonic echo signals of the weld seam for each detection cycle over a period of time and the phased array ultrasonic parameters in the machine learning processing platform; Based on each of the ultrasonic echo signals, each echo feature information reflecting the echo quality is determined; the echo feature information includes defect feature information, background scattering intensity information, echo morphology stability feature information, and parameter variation feature information. Based on the characteristic information of each echo, the change state of the echo quality is determined; wherein, the change state includes instantaneous fluctuation state and continuous deviation fluctuation state; Based on the changing state of the echo quality, a parameter adjustment scheme is determined; The phased array ultrasonic parameters are adjusted using the aforementioned parameter adjustment scheme to obtain the adjusted phased array ultrasonic parameters, thereby completing the adjustment of the phased array ultrasonic parameters.

[0006] Furthermore, the step of determining each echo characteristic information reflecting the echo quality based on each of the said ultrasonic echo signals includes: Each of the ultrasonic echo signals is filtered to obtain a filtered ultrasonic echo signal. The filtered ultrasonic echo signal is subjected to signal correction processing to obtain the corrected ultrasonic echo signal; Based on the corrected ultrasonic echo signal, each echo characteristic information reflecting the echo quality is determined.

[0007] Furthermore, the step of determining each echo characteristic information reflecting the echo quality based on the corrected ultrasonic echo signal includes: The corrected ultrasonic echo signal is subjected to signal feature extraction processing to obtain the mixed features reflecting the echo signal and the types and number of mixed features; Based on the mixed characteristics and the number of types of mixed characteristics that reflect the echo signal, information on each echo characteristic that reflects the echo quality is obtained.

[0008] Furthermore, the step of obtaining information about each echo feature reflecting the echo quality based on the mixing characteristics and the number of types of mixing characteristics of the echo signal includes: Based on the number of types of the hybrid features, a feature classification model is determined; Using the aforementioned feature classification model, the mixed features reflecting the echo signal are classified to obtain the echo feature information reflecting the echo quality.

[0009] Furthermore, the step of determining the change state of echo quality based on each echo characteristic information includes: The echo quality is evaluated using the echo feature information of each echo to obtain the feature evaluation value of the echo quality; The characteristic evaluation value of the echo quality is compared with the evaluation threshold to obtain the evaluation ratio; Based on the evaluation ratio, the change state of the echo quality is determined.

[0010] Furthermore, the step of determining the change state of echo quality based on the evaluation ratio includes: Based on the ultrasonic echo signal, the spectral distribution characteristics of the ultrasonic echo signal are determined; Based on the aforementioned spectral distribution characteristics and historical distribution characteristics, the influence weight of the echo quality change in the current period is determined; Based on the influence weight of the echo quality change in the current period and the evaluation ratio, the change state of the echo quality change is determined.

[0011] Further, the step of determining the change state of echo quality change based on the influence weight of the echo quality change in the current period and the evaluation ratio includes: The evaluation ratio is adjusted by using the influence weight of the echo quality change in the current period to obtain the adjusted evaluation ratio; The adjusted evaluation ratio is matched with a preset change state model to obtain the change state of echo quality.

[0012] Furthermore, the step of determining the parameter adjustment scheme based on the changing state of the echo quality includes: Based on the change state of the echo quality, the rate of change of the echo quality of the ultrasonic echo signal over a period of time is determined; The rate of change of the echo quality is matched with a preset adjustment scheme library to obtain a parameter adjustment scheme.

[0013] Further, the step of matching the rate of change of the echo quality with a preset adjustment scheme library to obtain a parameter adjustment scheme includes: The rate of change of the echo quality is matched with a preset adjustment scheme library to obtain all candidate matching schemes; All candidate matching schemes are filtered to obtain parameter adjustment schemes.

[0014] The present invention also provides a phased array ultrasonic parameter adjustment system based on machine learning, the system comprising: The signal acquisition module is used to acquire the ultrasonic echo signal of the weld in each detection cycle over a period of time and the phased array ultrasonic parameters in the machine learning processing platform; The feature determination module is used to determine each echo feature information reflecting the echo quality based on each of the ultrasonic echo signals; the echo feature information includes defect feature information, background scattering intensity information, echo morphology stability feature information, and parameter variation feature information; The state determination module is used to determine the change state of the echo quality change based on the characteristic information of each echo; wherein the change state includes instantaneous fluctuation state and continuous deviation fluctuation state; The scheme determination module is used to determine the parameter adjustment scheme based on the change state of the echo quality change; The scheme adjustment module is used to adjust the phased array ultrasonic parameters using the parameter adjustment scheme to obtain the adjusted phased array ultrasonic parameters, thereby completing the adjustment of the phased array ultrasonic parameters.

[0015] Compared with existing technologies, the machine learning-based phased array ultrasonic parameter adjustment method and system of the present invention have the following advantages: This invention acquires ultrasonic echo signals from each detection cycle of the weld over a period of time and phased array ultrasonic parameters from a machine learning processing platform. Based on each ultrasonic echo signal, it determines each echo feature reflecting the echo quality; based on each echo feature, it determines the state of change in echo quality; subsequently, based on the state of change in echo quality, it determines a parameter adjustment scheme and uses this scheme to adjust the phased array ultrasonic parameters, thus completing the adjustment of the phased array ultrasonic parameters. By introducing multi-dimensional echo feature information, including defect feature information, background scattering intensity information, stable echo morphology features, and parameter change features, it can more comprehensively and meticulously characterize the echo quality and distinguish between instantaneous fluctuation states and continuous deviation fluctuation states, thereby achieving decoupled attribution of independent parameter contributions. Furthermore, it can accurately determine the direction of parameter adjustment, effectively suppress periodic alternating swings of parameters outside the optimal range, significantly improve the detection accuracy of minute defects, and increase the accuracy of weld inspection. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention, the accompanying drawings used in the specific embodiments will be briefly described below. In all the drawings, the elements or parts are not necessarily drawn to scale.

[0017] Figure 1 This is a flowchart of a phased array ultrasonic parameter adjustment method based on machine learning according to the present invention.

[0018] Figure 2 This is a flowchart of step S200 in a phased array ultrasonic parameter adjustment method based on machine learning according to the present invention.

[0019] Figure 3 This is a flowchart of step S300 in a phased array ultrasonic parameter adjustment method based on machine learning according to the present invention.

[0020] Figure 4 This is a structural block diagram of a phased array ultrasonic parameter adjustment system based on machine learning according to the present invention.

[0021] In the diagram: 410, Signal Acquisition Module; 420, Feature Determination Module; 430, State Determination Module; 440, Scheme Determination Module; 450, Scheme Adjustment Module.

[0022] The implementation and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0023] The following drawings disclose several embodiments of the present invention. For clarity, many practical details will be described in the following description. However, it should be understood that these practical details are not intended to limit the invention. That is, in some embodiments of the invention, these practical details are not essential. Furthermore, for the sake of simplicity, some conventional structures and components will be shown in the drawings in a simple schematic manner.

[0024] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0025] Furthermore, in this invention, the use of terms such as "first" and "second" is for descriptive purposes only and does not specifically refer to any order or sequence, nor is it intended to limit the invention. They are merely used to distinguish components or operations described using the same technical terms, and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but only if they are feasible for those skilled in the art. If a combination of technical solutions is contradictory or impossible to implement, such a combination should be considered nonexistent and not within the scope of protection claimed by this invention.

[0026] To further understand the content, features, and effects of this invention, the following embodiments are provided, and detailed descriptions are given below in conjunction with the accompanying drawings: Please see Figure 1 This invention provides a method for adjusting phased array ultrasonic parameters based on machine learning, comprising the following steps: S100. Acquire the ultrasonic echo signals of the weld seam for each inspection cycle over a period of time and the phased array ultrasonic parameters from the machine learning processing platform. The phased array ultrasonic parameters refer to the key adjustable parameters affecting the formation, propagation, and reception of the ultrasonic beam during phased array ultrasonic testing, such as the emission angle, center frequency, receiving gain, focusing depth, and pulse width; the combination of these parameters determines the performance of the ultrasonic testing and the quality of the echo signal. The ultrasonic echo signal is the sound wave signal reflected back when ultrasonic waves penetrate the workpiece and encounter an interface or defect; it contains important information about the internal structure and defects of the workpiece. The machine learning processing platform refers to the hardware and software environment integrating machine learning algorithms and computing resources, used to process ultrasonic data, train models, perform decision-making inference, and ultimately output parameter adjustment instructions. Specifically, the ultrasonic echo signals are acquired in real time through phased array ultrasonic equipment. For example, during weld seam inspection, the ultrasonic probe emits ultrasonic waves at a preset scanning path and frequency and receives the ultrasonic signals reflected back from inside the weld seam; these signals can be raw analog signals or digital signals after analog-to-digital conversion. The phased array ultrasound parameters can be obtained from the machine learning processing platform. These parameters are usually the parameter settings used in the current detection cycle, such as the current transmission angle, center frequency, and receiving gain.

[0027] S200. Based on each of the ultrasonic echo signals, determine each echo characteristic information reflecting the echo quality; the echo characteristic information includes defect characteristic information, background scattering intensity information, stationary echo morphology characteristic information, and parameter variation characteristic information. Here, echo characteristic information refers to indicators extracted from the ultrasonic echo signal that can quantify the echo quality and defect characteristics, such as defect characteristic information, background scattering intensity information, stationary echo morphology characteristic information, and parameter variation characteristic information. Specifically, time-domain or frequency-domain analysis can be performed on each ultrasonic echo signal to extract its amplitude, phase, frequency, bandwidth, and other features. Defect characteristic information can include the amplitude, duration, location, and shape of the defect echo, which can be extracted from the echo signal using signal processing algorithms (such as envelope detection, peak detection, etc.). Background scattering intensity information reflects the degree of scattering of ultrasonic waves propagating in a material, and can be obtained, for example, by calculating the root mean square value or energy of the echo signal in a defect-free region. Stationary echo morphology characteristic information can assess the stability of the echo signal, and can be obtained, for example, by calculating the variance, standard deviation, or autocorrelation function of the echo signal. Parameter change characteristic information can reflect the changes in parameter settings between the current detection cycle and historical detection cycles. For example, it can be obtained by comparing the difference or rate of change between the current parameter and the parameter of the previous cycle.

[0028] S300. Based on each echo feature information, determine the change state of the echo quality; wherein, the change state includes instantaneous fluctuation state and continuous deviation fluctuation state. Specifically, it is determined according to the numerical change trend of the echo feature information. If an echo feature information fluctuates drastically in a short period of time but quickly returns to normal, it can be judged as an instantaneous fluctuation state. If an echo feature information continuously deviates from the normal range, and the degree of deviation gradually increases, it can be judged as a continuous deviation fluctuation state. This can be achieved by setting a threshold and a time window. For example, when a feature value exceeds a preset threshold for multiple consecutive periods, it is determined as a continuous deviation fluctuation state.

[0029] S400. Based on the changing state of the echo quality, determine a parameter adjustment scheme. Specifically, if the echo quality is in a state of instantaneous fluctuation, small-scale fine-tuning may be necessary to avoid overreaction. If the echo quality is in a state of continuous deviation from fluctuation, larger-scale adjustments may be necessary, and even adjustments to multiple parameters may be required. The parameter adjustment scheme can be a preset set of rules or a decision strategy trained on historical data by a machine learning model (such as a reinforcement learning model). For example, when a persistently high background scattering intensity is detected, the parameter adjustment scheme may suggest reducing the center frequency or adjusting the emission angle to reduce the scattering effect.

[0030] S500. Using the aforementioned parameter adjustment scheme, the phased array ultrasonic parameters are adjusted to obtain the adjusted phased array ultrasonic parameters, thereby completing the adjustment of the phased array ultrasonic parameters. Specifically, according to the determined parameter adjustment scheme, the machine learning processing platform generates specific parameter adjustment instructions and sends them to the phased array ultrasonic equipment via an interface. After receiving the instructions, the ultrasonic equipment adjusts its transmission angle, center frequency, receiving gain, and other parameters accordingly, thereby optimizing the ultrasonic echo quality for the next detection cycle.

[0031] In this embodiment, the ultrasonic echo signal of the weld and the current phased array ultrasonic parameters are acquired. Each ultrasonic echo signal is then finely processed to extract multiple dimensions of echo feature information, including defect characteristics, background scattering intensity, echo morphology stability, and parameter variation. This feature information provides a more comprehensive and detailed reflection of the current echo quality, avoiding the limitations of a single indicator. Next, based on the multi-dimensional echo feature information, the state of echo quality change is comprehensively judged, distinguishing between instantaneous fluctuations and continuous deviations. Subsequently, based on the state of echo quality change, the optimal parameter adjustment scheme is intelligently determined. This scheme is no longer a simple linear feedback but is generated based on a deep understanding of echo quality changes. Finally, using the determined parameter adjustment scheme, the phased array ultrasonic parameters are adjusted in real time. This accurately determines the direction of parameter adjustment, thereby optimizing the ultrasonic detection effect, effectively suppressing periodic alternating swings of parameters outside the optimal range, significantly improving the detection accuracy of minute defects, and increasing the accuracy of weld inspection.

[0032] Please refer to some of the embodiments described above in this application. Figure 2 Step S200: The step of determining each echo characteristic information reflecting the echo quality based on each of the said ultrasonic echo signals includes: S210. Each of the ultrasonic echo signals is filtered to obtain a filtered ultrasonic echo signal. Filtering each ultrasonic echo signal removes random noise, high-frequency interference, or other irrelevant components to improve the signal-to-noise ratio. Various techniques can be used, such as low-pass filtering, band-pass filtering, median filtering, or wavelet denoising, depending on the characteristics of the noise and the signal requirements. Filtering results in a cleaner filtered ultrasonic echo signal.

[0033] S220. Perform signal correction processing on the filtered ultrasonic echo signal to obtain a corrected ultrasonic echo signal. The purpose of this step is to compensate for attenuation, distortion, or drift that may occur during signal transmission and reception, ensuring the accuracy of the signal's amplitude, phase, and time reference. Signal correction processing may include baseline drift correction, amplitude normalization, time gain compensation (TGC), or phase correction. For example, in ultrasonic testing, as the sound wave travels further in the medium, its energy gradually attenuates. Time gain compensation can compensate for the gain of long-distance echoes, making their amplitude comparable to that of short-distance echoes. Thus, a corrected ultrasonic echo signal can be obtained.

[0034] S230. Based on the corrected ultrasonic echo signal, determine each echo characteristic information reflecting the echo quality. Specifically, determining each echo characteristic information reflecting the echo quality based on the corrected ultrasonic echo signal allows for more accurate determination. These echo characteristic information, such as defect characteristic information, background scattering intensity information, echo morphology stability characteristic information, and parameter variation characteristic information, will more realistically reflect the internal state of the weld and the performance of the detection system.

[0035] This invention effectively avoids potential noise interference and signal distortion in the original ultrasonic echo signal by introducing filtering and signal correction processing before determining echo feature information. Filtering effectively suppresses random noise and high-frequency clutter, making the effective components of the signal more prominent and providing high-quality input for subsequent feature extraction. Signal correction further compensates for attenuation and distortion during signal propagation and reception, ensuring signal authenticity and consistency. Because of this preprocessing of the ultrasonic echo signal, the echo feature information determined based on the corrected ultrasonic echo signal can more accurately and reliably reflect the actual condition of the weld and the echo quality, thus providing a better data foundation for machine learning processing platforms.

[0036] In some embodiments of this application described above, the step of determining each echo characteristic information reflecting the echo quality based on the corrected ultrasonic echo signal includes: The corrected ultrasonic echo signal undergoes signal feature extraction processing to obtain mixed features reflecting the echo signal and the number of types of mixed features. Signal feature extraction processing refers to identifying and quantifying various attributes related to echo quality from the corrected ultrasonic echo signal. These attributes may include, but are not limited to, signal amplitude, frequency, phase, energy distribution, time-domain waveform characteristics, and frequency-domain spectral characteristics. By transforming, analyzing, and calculating the original signal, a series of numerical values ​​or vectors characterizing the signal properties can be obtained, i.e., mixed features. The number of mixed feature types indicates the types or dimensions of the extracted features. For example, multiple features such as peak amplitude, rise time, fall time, pulse width, center frequency, bandwidth, energy, and signal-to-noise ratio can be extracted; each feature represents a type.

[0037] Based on the mixed features reflecting the echo signal and the number of types of mixed features, information on each echo feature reflecting the echo quality is obtained. Specifically, through weighted averaging, machine learning model input, rule-based reasoning, and other methods, multiple mixed features and their types and numbers are integrated to obtain more comprehensive and accurate information on defect features, background scattering intensity, echo morphology stability, and parameter variation.

[0038] This invention employs refined signal feature extraction processing on corrected ultrasonic echo signals to identify multi-dimensional and multi-type mixed features from complex ultrasonic echoes. These mixed features contain rich information about the ultrasonic echo signal, such as signal intensity, shape, and frequency components. Through comprehensive analysis of the mixed features and their types and quantities, the quality of the ultrasonic echo can be more comprehensively and accurately characterized. The step-by-step and meticulous feature extraction and integration process allows subsequent echo quality assessments to be based on a more solid and representative data foundation, thereby improving the accuracy and reliability of echo feature information.

[0039] In some embodiments of this application described above, the step of obtaining information about each echo feature reflecting the echo quality based on the mixing characteristics reflecting the echo signal and the number of types of mixing characteristics includes: Based on the number of types of the mixed features, a feature classification model is determined. The number of types of mixed features refers to the variety or dimensionality of the mixed features obtained after signal feature extraction processing of the corrected ultrasonic echo signal. This number of types reflects the different properties or sources of feature components contained in the ultrasonic echo signal. Specifically, determining the feature classification model means selecting or constructing a suitable machine learning classification model based on the number of types of mixed features. For example, when the number of mixed feature types is small and the features are linearly separable, a support vector machine or logistic regression model can be used; when the features are complex and the nonlinear relationships are significant, models such as neural networks, decision trees, or random forests can be used. The feature classification model aims to effectively distinguish complex mixed features into different echo feature information, such as defect feature information, background scattering intensity information, stationary echo morphology feature information, and parameter variation feature information.

[0040] Using the aforementioned feature classification model, the mixed features reflecting the echo signal are classified to obtain individual echo feature information reflecting the echo quality. This step involves inputting the extracted mixed features into the established feature classification model. The model's classification algorithm maps or categorizes the mixed features into predefined echo feature information categories. Thus, various feature information directly related to echo quality can be accurately identified and extracted from complex raw signals.

[0041] This invention determines a suitable feature classification model based on the number of types of mixed features, ensuring that the classification model can effectively adapt to and handle the feature complexity contained in current ultrasonic echo signals. Subsequently, the feature classification model is used to classify the mixed features reflecting the echo signal, accurately decomposing and classifying the original, potentially mixed features into echo feature information with clear physical meaning. This avoids making fuzzy judgments directly from mixed features, improving the accuracy and reliability of feature extraction.

[0042] For some of the embodiments described above in this application, please refer to Figure 3 Step S300: The step of determining the change state of echo quality based on the echo characteristic information of each echo includes: S310. Using each echo feature information, the echo quality is evaluated to obtain a feature evaluation value for the echo quality. Specifically, the echo feature information includes defect feature information, background scattering intensity information, echo morphology stability feature information, and parameter variation feature information. These feature information can be comprehensively considered when evaluating the echo quality. For example, defect feature information can reflect the existence or type of defects; background scattering intensity information can reflect the attenuation and scattering of the ultrasonic signal in the medium; echo morphology stability feature information can reflect the signal stability; and parameter variation feature information can reflect the changing trend of the phased array ultrasonic parameters themselves. By weighting these features or performing comprehensive analysis through a machine learning model, a quantified echo quality feature evaluation value can be obtained, which comprehensively reflects the current quality status of the ultrasonic echo signal.

[0043] S320. The characteristic evaluation value of the echo quality is compared with the evaluation threshold to obtain an evaluation ratio. The evaluation threshold is a pre-set standard value or set of standard values ​​used to measure whether the characteristic evaluation value of the echo quality is within the normal range or an acceptable fluctuation range. This evaluation threshold can be set based on historical data, expert experience, or the requirements of a specific application scenario. Comparing the characteristic evaluation value of the echo quality with the evaluation threshold yields the evaluation ratio, which intuitively reflects the degree of deviation of the current echo quality from the preset standard. For example, the evaluation ratio can be a percentage, indicating how much deviation has occurred.

[0044] S330. Based on the evaluation ratio, determine the state of change in echo quality. This step refers to determining whether the echo quality is in a state of instantaneous fluctuation or a state of continuous deviation from fluctuation based on the magnitude and trend of the evaluation ratio. For example, if the evaluation ratio fluctuates slightly in a short period of time, it may be judged as a state of instantaneous fluctuation; if the evaluation ratio continuously exceeds a certain range for a long period of time, it may be judged as a state of continuous deviation from fluctuation. This quantification and comparison method makes the judgment of echo quality changes more objective and accurate.

[0045] Specifically, during weld inspection, a series of ultrasonic echo signals are acquired. First, using defect feature information, background scattering intensity information, echo morphology stability information, and parameter variation information from the echo signals, a pre-trained machine learning model (e.g., support vector machine or neural network) comprehensively evaluates the echo quality of the current cycle, obtaining a feature evaluation value between 0 and 100, where a higher value indicates better echo quality. For example, an evaluation value of 85. Next, this feature evaluation value is compared with a preset evaluation threshold, for example, a normal fluctuation threshold of 80-95. If the evaluation value of 85 is within this range, an evaluation ratio is calculated. If the evaluation value is below 80 or above 95, the evaluation ratio indicates a deviation. Based on the evaluation ratio, it is determined whether the current echo quality is in a state of instantaneous fluctuation (e.g., the evaluation value fluctuates slightly between 80 and 95) or continuously deviates from the fluctuation state (e.g., the evaluation value is below 80 or above 95 for several consecutive cycles). For example, if the evaluation ratio shows that the echo quality is below 80 for three consecutive cycles, it is judged as a continuous deviation from the fluctuation state, thereby triggering a more aggressive parameter adjustment strategy.

[0046] This invention transforms abstract echo characteristic information into quantified characteristic evaluation values ​​and evaluation ratios by introducing an evaluation process for echo quality and comparing it with an evaluation threshold. Due to this quantification and comparison mechanism, the actual state of echo quality and its degree of deviation can be identified more accurately. Through the evaluation ratio, changes in echo quality can be effectively divided into instantaneous fluctuation states and continuous deviation fluctuation states, thus providing a more refined and accurate basis for subsequent parameter adjustments and avoiding errors that may arise from subjective judgments based solely on characteristic information.

[0047] In some embodiments of this application described above, the step of determining the change state of echo quality based on the evaluation ratio includes: Based on the ultrasonic echo signal, the spectral distribution characteristics of the ultrasonic echo signal are determined. The spectral distribution characteristics refer to the energy distribution of the ultrasonic echo signal at different frequencies, reflecting information such as the signal's frequency components, energy concentration, and the presence of abnormal frequency components. For example, the spectral distribution characteristics can be obtained by performing spectral analysis methods such as Fourier transform on the ultrasonic echo signal.

[0048] Based on the aforementioned spectral distribution characteristics and historical distribution characteristics, the influence weight of the echo quality change in the current period is determined. Specifically, the historical distribution characteristics refer to the statistical regularity or trend of the spectral distribution characteristics of the ultrasonic echo signal over a past period, which can serve as a reference benchmark for judging whether the current echo signal is abnormal. The influence weight represents the importance or correction factor for judging the current echo quality change, and its purpose is to correct or enhance judgments based solely on the evaluation ratio, according to the inherent physical characteristics and historical evolution patterns of the ultrasonic echo signal. For example, if the current spectral distribution characteristics are highly consistent with the historical distribution characteristics, even if the evaluation ratio fluctuates slightly, the influence weight may indicate that it is a normal fluctuation; conversely, if the spectral distribution characteristics deviate significantly, the influence weight will increase the tendency to judge an abnormal state.

[0049] Based on the influence weight of the echo quality change in the current period and the evaluation ratio, the change state of the echo quality change is determined. Specifically, the evaluation ratio provides a preliminary quantitative judgment, while the influence weight introduces deeper physical and time series information of the signal, making the determination of the change state more comprehensive and reliable.

[0050] Specifically, during a weld inspection, an ultrasonic echo signal is acquired. First, a Fourier transform is performed on the ultrasonic echo signal to obtain its spectral distribution characteristics. For example, it is found that the dominant frequency energy is concentrated at 5MHz, and the bandwidth is 1MHz. Simultaneously, historical data is reviewed, revealing that over the past 100 inspection cycles, the dominant frequency of the ultrasonic echo signal for this type of weld is typically between 4.8MHz and 5.2MHz, and the bandwidth is between 0.9MHz and 1.1MHz, with a stable spectral pattern. At this point, the spectral distribution characteristics of the current cycle highly match the historical distribution characteristics, indicating that the physical characteristics of the signal are within the normal range. Based on this high degree of consistency, the system determines a high influence weight, for example, 0.9. Even if the evaluation ratio obtained by comparing the feature evaluation value of the echo quality with the evaluation threshold slightly exceeds the normal range (e.g., the evaluation ratio is 1.1, while the threshold is 1.0), due to the high influence weight, this deviation is considered to be a normal or transient fluctuation, rather than a continuous deviation. For example, it is ultimately determined to be a transient fluctuation. Conversely, if the current spectral distribution of the ultrasonic echo signal shows a dominant frequency shift to 6 MHz and a significantly narrower bandwidth, which differs markedly from historical distribution characteristics, a lower influence weight, such as 0.3, is assigned. Even if the evaluation ratio deviates only slightly (e.g., 1.05), the lower influence weight makes it more likely to be judged as a sustained deviation, as significant spectral changes may indicate deeper quality issues or changes in the detection environment. This influence weighting mechanism makes the judgment of echo quality changes more intelligent and contextualized, avoiding rigid threshold judgments and improving the system's adaptability and accuracy.

[0051] This invention optimizes the process of determining the changing state of echo quality by introducing the spectral and historical distribution characteristics of ultrasonic echo signals and determining the influence weights accordingly. Specifically, the spectral distribution characteristics can reveal subtle changes in the ultrasonic echo signal in the frequency domain. These changes are closely related to factors such as the internal structure of the weld, defect type, and noise interference, and are difficult to fully capture by simply measuring the signal strength or morphology in the time domain. Simultaneously, the introduction of historical distribution characteristics can establish a dynamic benchmark, distinguishing between normal system fluctuations and true quality deviations. By combining this multi-dimensional information and correcting or weighting the evaluation ratio through influence weights, a deeper understanding of the signal's essence can be achieved when judging echo quality changes, avoiding misjudgments due to instantaneous fluctuations of a single indicator, or promptly detecting potential, slowly developing quality problems.

[0052] In some embodiments of this application described above, the step of determining the change state of echo quality change based on the influence weight of the echo quality change in the current period and the evaluation ratio includes: The evaluation ratio is adjusted using the influence weight of the echo quality change in the current period to obtain an adjusted evaluation ratio. Specifically, the adjustment process refers to modifying or weighting the original evaluation ratio according to the influence weight of the echo quality change in the current period. For example, a higher influence weight may indicate an increased specialness or importance of the current echo signal. In this case, the evaluation ratio can be amplified or reduced to better reflect the actual situation in the current period. The aim is to enable the evaluation ratio to more accurately reflect the actual deviation of the current echo quality, especially when specific influencing factors are present.

[0053] The adjusted evaluation ratio is matched with a preset change state model to obtain the change state of echo quality. The preset change state model is a pre-established classification or regression model used to identify different echo quality change states. This model can be trained on a large amount of historical data to learn the mapping relationship between different adjusted evaluation ratios and actual echo quality change states (e.g., instantaneous fluctuation state, continuous deviation fluctuation state, etc.). This model can be a decision tree, support vector machine, neural network, or other machine learning model, or it can be a lookup table containing different thresholds and rules. Its purpose is to provide a standardized judgment criterion, mapping the adjusted evaluation ratio to a specific echo quality change state. The matching process specifically refers to inputting the adjusted evaluation ratio into the preset change state model, and the model outputting the corresponding echo quality change state. For example, if the model is a decision tree, the adjusted evaluation ratio will be judged along the path of the decision tree, eventually falling into a leaf node, which represents a change state. Its purpose is to utilize the model's intelligent judgment capability to transform the quantified evaluation ratio into a meaningful description of the change state.

[0054] Specifically, within a detection cycle, the spectral distribution characteristics of the ultrasonic echo signal indicate the presence of high-frequency noise interference, resulting in an impact weight of 0.8 (indicating high impact) for the echo quality change in the current cycle. Simultaneously, the evaluation ratio obtained through processing is 1.2 (indicating a slight decrease in echo quality). Directly using an evaluation ratio of 1.2 might be interpreted as a slight fluctuation. However, considering the impact weight of 0.8, this invention adjusts the evaluation ratio using this weight. For example, a weighted average method can be used, or the evaluation ratio can be corrected according to a preset rule to obtain an adjusted evaluation ratio, such as 1.5. Subsequently, this adjusted evaluation ratio of 1.5 is input into a preset change state model. This model may be a neural network trained on historical data, capable of identifying that when the impact weight is high and the evaluation ratio reaches 1.5, the current echo quality change is more likely to be a momentary fluctuation rather than a continuous deviation from the fluctuation state, or vice versa, providing a more accurate judgment based on the specific training results of the model. In this way, even if the original evaluation ratio does not change much, by combining the influence weights for adjustment and model matching, the true changes in echo quality can be identified more accurately.

[0055] This invention adjusts the evaluation ratio by introducing influence weights, allowing it to better adapt to the specific conditions of the current detection cycle and avoiding potential biases from directly using the original evaluation ratio. This adjustment enables the evaluation ratio to more accurately reflect the true deviation of the current echo quality. Furthermore, the adjusted evaluation ratio is matched with a preset change state model. Leveraging the model's learning and generalization capabilities, it can more effectively identify complex and variable echo quality changes, such as distinguishing between transient noise interference and persistent defect signal changes, thereby improving the accuracy and robustness of state judgment.

[0056] In some embodiments of this application described above, the step of determining a parameter adjustment scheme based on the changing state of the echo quality includes: Based on the changing state of the echo quality, the rate of change of the echo quality of the ultrasonic echo signal over a period of time is determined. Specifically, the rate of change of echo quality refers to the speed or trend of the echo quality of the ultrasonic echo signal transitioning from one state of change to another within multiple consecutive detection cycles, or the degree of deviation from the normal range over time within the same state of change. This rate of change can quantify the dynamic process of echo quality deterioration or improvement; for example, it can be expressed as the average decrease or increase in the echo quality assessment value per unit time, or as the time required for a momentary fluctuation to develop into a sustained deviation from the fluctuation state. The aim is to provide more dynamic and quantitative information than a single state of change, in order to provide more precise guidance for parameter adjustments.

[0057] The rate of change of the echo quality is matched with a pre-defined adjustment scheme library to obtain parameter adjustment schemes. This library is a database storing various parameter adjustment strategies. Each adjustment scheme in the library is associated with a specific range or pattern of echo quality change rate. For example, when the rate of change indicates a rapid deterioration in echo quality, an aggressive adjustment scheme might be corresponding to it; conversely, when the rate of change indicates a slow decline in echo quality, a milder adjustment scheme might be appropriate. The adjustment scheme library can be pre-trained using expert experience, historical data analysis, or machine learning models, aiming to provide a systematic and standardized adjustment strategy to address various possible echo quality changes. The process involves comparing the currently detected rate of change of echo quality with the range or pattern of change corresponding to each scheme stored in the adjustment scheme library. Various algorithms can be used in the matching process, such as rule-based matching, fuzzy matching, or machine learning classifier matching. Through matching, the parameter adjustment scheme that best matches the current echo quality change trend and degree can be selected from the library. The purpose is to ensure that the selected adjustment scheme can accurately respond to the current echo quality problem, thereby achieving efficient and effective parameter adjustment.

[0058] Specifically, if the echo quality assessment value of the continuously detected ultrasonic echo signal consistently decreases over a period of time and is identified as being in a state of continuous deviation and fluctuation, then the average rate of decrease of this echo quality assessment value over the past five detection cycles is further calculated. For example, if the rate of change of the echo quality is found to be 0.5 units per cycle, then this rate of change (0.5 units per cycle) is input into a preset adjustment scheme library for matching. The adjustment scheme library may have preset rules such as: if the rate of change is between 0.3 and 0.6 units per cycle, then a fine-tuning gain scheme is recommended, specifically increasing the transmission gain in the phased array ultrasonic parameters by 2 dB and simultaneously fine-tuning the depth of focus by 0.5 mm. Through matching processing, the system can automatically select and execute this fine-tuning gain scheme to adjust the phased array ultrasonic parameters in order to improve the echo quality.

[0059] This invention, by introducing the rate of change of echo quality, enables a deeper analysis of the dynamic evolution of echo quality, overcoming the limitations of judging based on a single change state. Quantitative analysis of the rate of change allows the system to capture subtle trends and speeds of echo quality changes, providing richer and more precise data for subsequent parameter adjustments. Based on this, by matching the rate of change with a pre-set adjustment scheme library, the system can systematically select the most suitable parameter adjustment scheme for the current echo quality change. This ensures the timeliness, relevance, and effectiveness of parameter adjustments, avoiding over-adjustment or under-adjustment due to insufficient information.

[0060] In some embodiments of this application described above, the step of matching the rate of change of the echo quality with a preset adjustment scheme library to obtain a parameter adjustment scheme includes: The rate of change of the echo quality is matched against a preset adjustment scheme library to obtain all candidate matching schemes. This step means that a certain degree of ambiguity or similarity is allowed during the matching process, thereby identifying all adjustment schemes that match the rate of change of the current echo quality within a preset threshold range. There may be more than one candidate matching scheme, representing multiple potentially effective response strategies to the current echo quality change. The preset adjustment scheme library can be a database containing various parameter adjustment strategies and their corresponding triggering conditions or rate of change ranges.

[0061] The process involves filtering all candidate matching schemes to obtain parameter adjustment schemes. This step involves selecting, based on additional judgment criteria or optimization objectives, the single or group of parameter adjustment schemes that best suit the current operating conditions or deliver the optimal results. The filtering process can be based on various strategies, such as combining historical data analysis, real-time operating parameters, expert experience rules, or further evaluating and ranking the candidate schemes using machine learning models. The aim is to improve the accuracy, effectiveness, and adaptability of the final selected adjustment scheme.

[0062] Specifically, if the rate of change in the detected echo quality is at a moderately high level within a detection cycle, it indicates the presence of some degree of defect within the weld or a change in the detection environment. When this rate of change is matched against a pre-defined adjustment scheme library, multiple candidate matching schemes may be identified, such as: Scheme A: slightly increase the ultrasonic transmission power and adjust the focusing depth; Scheme B: change the ultrasonic frequency and perform gain compensation; Scheme C: adjust the probe incident angle and reduce the scanning speed. The selection process can then be implemented as follows: First, historical data is queried, revealing that for the current weld material and thickness, Scheme A typically achieves the best defect identification effect at a similar rate of change. Second, considering real-time operating conditions, it is found that the current welding speed is relatively fast, and Scheme C may significantly reduce detection efficiency. Finally, a pre-trained machine learning model is used, inputting all current operating parameters and the features of the three candidate schemes, and the model outputs a score for the expected effect of each scheme. Based on a comprehensive evaluation, Scheme A is ultimately selected as the parameter adjustment scheme.

[0063] This invention effectively addresses the limitations of direct matching by introducing a step to obtain all candidate matching schemes and then performing a screening process. When the rate of change in echo quality matches multiple schemes in the adjustment scheme library to a certain extent, instead of simply selecting the first or any matching item, all potentially effective schemes are considered. Subsequently, through screening, more dimensions of information can be comprehensively considered, such as real-time operating parameters like the current weld material type, thickness, welding speed, and ambient temperature, or combined with historical adjustment effect data, to conduct a refined evaluation of these candidate schemes. This ensures that the final selected parameter adjustment scheme not only matches the rate of change but is also the optimal or most suitable scheme under the current conditions, thus avoiding suboptimal or inaccurate adjustments that may result from simple matching.

[0064] For a machine learning-based phased array ultrasonic parameter adjustment method based on any of the above embodiments, please refer to [link to relevant documentation]. Figure 4The present invention also provides a phased array ultrasonic parameter adjustment system based on machine learning, the system comprising a signal acquisition module 410, a feature determination module 420, a state determination module 430, a scheme determination module 440, and a scheme adjustment module 450.

[0065] The signal acquisition module 410 is used to acquire the ultrasonic echo signal of the weld in each detection cycle over a period of time and the phased array ultrasonic parameters in the machine learning processing platform.

[0066] The feature determination module 420 is used to determine each echo feature information reflecting the echo quality based on each of the ultrasonic echo signals; the echo feature information includes defect feature information, background scattering intensity information, echo morphology stability feature information, and parameter variation feature information.

[0067] The state determination module 430 is used to determine the change state of the echo quality change based on the characteristic information of each echo; wherein the change state includes instantaneous fluctuation state and continuous deviation fluctuation state.

[0068] The scheme determination module 440 is used to determine the parameter adjustment scheme based on the change state of the echo quality change.

[0069] The scheme adjustment module 450 is used to adjust the phased array ultrasonic parameters using the parameter adjustment scheme to obtain the adjusted phased array ultrasonic parameters, thereby completing the adjustment of the phased array ultrasonic parameters.

[0070] In this embodiment, the signal acquisition module 410 acquires ultrasonic echo signals and current phased array ultrasonic parameters in real time, and the feature determination module 420 performs multi-dimensional and refined feature extraction on the echo signals. Subsequently, the state determination module 430 accurately determines the change state of the echo quality based on rich feature information, distinguishing between instantaneous fluctuations and continuous deviations. On this basis, the scheme determination module 440 can generate targeted parameter adjustment schemes, and finally, the scheme adjustment module 450 optimizes and adjusts the phased array ultrasonic parameters. This enables precise and adaptive adjustment of the phased array ultrasonic parameters, effectively improving the accuracy and stability of the detection.

[0071] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the present invention specification.

Claims

1. A method for adjusting phased array ultrasonic parameters based on machine learning, characterized in that, include: Acquire the ultrasonic echo signals of the weld seam for each detection cycle over a period of time and the phased array ultrasonic parameters in the machine learning processing platform; Based on each of the ultrasonic echo signals, each echo feature information reflecting the echo quality is determined; the echo feature information includes defect feature information, background scattering intensity information, echo morphology stability feature information, and parameter variation feature information. Based on the characteristic information of each echo, the change state of the echo quality is determined; wherein, the change state includes instantaneous fluctuation state and continuous deviation fluctuation state; Based on the changing state of the echo quality, a parameter adjustment scheme is determined; The phased array ultrasonic parameters are adjusted using the aforementioned parameter adjustment scheme to obtain the adjusted phased array ultrasonic parameters, thereby completing the adjustment of the phased array ultrasonic parameters.

2. The method for adjusting phased array ultrasonic parameters based on machine learning according to claim 1, characterized in that, The step of determining each echo characteristic information reflecting the echo quality based on each of the said ultrasonic echo signals includes: Each of the ultrasonic echo signals is filtered to obtain a filtered ultrasonic echo signal. The filtered ultrasonic echo signal is subjected to signal correction processing to obtain the corrected ultrasonic echo signal; Based on the corrected ultrasonic echo signal, each echo characteristic information reflecting the echo quality is determined.

3. The method for adjusting phased array ultrasonic parameters based on machine learning according to claim 2, characterized in that, The step of determining each echo characteristic information reflecting the echo quality based on the corrected ultrasonic echo signal includes: The corrected ultrasonic echo signal is subjected to signal feature extraction processing to obtain the mixed features reflecting the echo signal and the types and number of mixed features; Based on the mixed characteristics and the number of types of mixed characteristics that reflect the echo signal, information on each echo characteristic that reflects the echo quality is obtained.

4. The method for adjusting phased array ultrasonic parameters based on machine learning according to claim 3, characterized in that, The step of obtaining information about each echo feature reflecting the echo quality based on the mixing characteristics and the number of types of mixing features of the echo signal includes: Based on the number of types of the hybrid features, a feature classification model is determined; Using the aforementioned feature classification model, the mixed features reflecting the echo signal are classified to obtain the echo feature information reflecting the echo quality.

5. The method for adjusting phased array ultrasonic parameters based on machine learning according to claim 1, characterized in that, The steps for determining the change state of echo quality based on each echo characteristic information include: The echo quality is evaluated using the echo feature information of each echo to obtain the feature evaluation value of the echo quality; The characteristic evaluation value of the echo quality is compared with the evaluation threshold to obtain the evaluation ratio; Based on the evaluation ratio, the change state of the echo quality is determined.

6. The method for adjusting phased array ultrasonic parameters based on machine learning according to claim 5, characterized in that, Based on the evaluation ratio, the steps for determining the change state of echo quality include: Based on the ultrasonic echo signal, the spectral distribution characteristics of the ultrasonic echo signal are determined; Based on the aforementioned spectral distribution characteristics and historical distribution characteristics, the influence weight of the echo quality change in the current period is determined; Based on the influence weight of the echo quality change in the current period and the evaluation ratio, the change state of the echo quality change is determined.

7. The method for adjusting phased array ultrasonic parameters based on machine learning according to claim 6, characterized in that, The step of determining the change state of echo quality based on the influence weight of the echo quality change in the current period and the evaluation ratio includes: The evaluation ratio is adjusted by using the influence weight of the echo quality change in the current period to obtain the adjusted evaluation ratio; The adjusted evaluation ratio is matched with a preset change state model to obtain the change state of echo quality.

8. The method for adjusting phased array ultrasonic parameters based on machine learning according to claim 1, characterized in that, The steps for determining the parameter adjustment scheme based on the changes in echo quality include: Based on the change state of the echo quality, the rate of change of the echo quality of the ultrasonic echo signal over a period of time is determined; The rate of change of the echo quality is matched with a preset adjustment scheme library to obtain a parameter adjustment scheme.

9. The method for adjusting phased array ultrasonic parameters based on machine learning according to claim 8, characterized in that, The steps of matching the rate of change of the echo quality with a preset adjustment scheme library to obtain the parameter adjustment scheme include: The rate of change of the echo quality is matched with a preset adjustment scheme library to obtain all candidate matching schemes; All candidate matching schemes are filtered to obtain parameter adjustment schemes.

10. A phased array ultrasonic parameter adjustment system based on machine learning, characterized in that, The system includes: The signal acquisition module is used to acquire the ultrasonic echo signal of the weld in each detection cycle over a period of time and the phased array ultrasonic parameters in the machine learning processing platform; The feature determination module is used to determine each echo feature information reflecting the echo quality based on each of the ultrasonic echo signals; the echo feature information includes defect feature information, background scattering intensity information, echo morphology stability feature information, and parameter variation feature information; The state determination module is used to determine the change state of the echo quality change based on the characteristic information of each echo; wherein the change state includes instantaneous fluctuation state and continuous deviation fluctuation state; The scheme determination module is used to determine the parameter adjustment scheme based on the change state of the echo quality change; The scheme adjustment module is used to adjust the phased array ultrasonic parameters using the parameter adjustment scheme to obtain the adjusted phased array ultrasonic parameters, thereby completing the adjustment of the phased array ultrasonic parameters.

Images