Ultrasonic detection and data analysis method for impurity content of concentrated solution

Abstract

The invention relates to the technical field of detection, and discloses a concentrated solution impurity content ultrasonic detection and data analysis method and system. The method comprises the following steps: collecting multi-frequency-point echo signals in a frequency band of 500kHz-10MHz; extracting five types of ultrasonic features through time-frequency joint domain decomposition; after principal component analysis dimensionality reduction, inputting a particle size segmentation identification model, and calling a nonlinear mapping function of a corresponding interval to calculate the impurity concentration of each particle size; and in combination with a dynamic baseline correction mechanism based on mahalanobis distance, a result is corrected in real time according to viscosity and temperature. The system comprises a signal acquisition unit, a feature extraction unit, a dimension reduction unit, an identification unit, a concentration calculation unit, a baseline correction unit and an output unit. According to the method, the impurity quality inspection limit, the particle size recognition accuracy and the total content measurement precision are improved.

Description

Technical Field

[0001] This invention belongs to the field of detection technology, specifically relating to ultrasonic detection and data analysis methods for impurity content in concentrated solutions. Background Technology

[0002] With the increasing demands for purity in liquid media from industrial process control and product quality supervision, high-precision detection of impurities in concentrated solutions has become a core requirement in key fields such as pharmaceuticals, semiconductor manufacturing, and high-end chemicals. Ultrasonic testing technology, due to its non-invasiveness, real-time performance, and applicability to closed systems, is widely used in liquid composition analysis. Traditional ultrasonic testing methods primarily rely on single characteristic parameters for impurity identification, making it difficult to decouple the complex scattering and absorption signals generated by the bulk matrix of the concentrated solution. This leads to a decrease in the signal-to-noise ratio at low impurity concentrations or in the presence of micro- or nano-sized particles, easily resulting in missed detections or false positives.

[0003] Impurity content analysis based on ultrasonic signals focuses on extracting quantifiable indicators related to the physical properties of impurities from the received echoes. Ideally, this technique should be able to distinguish impurity particles of different size scales and establish a mapping relationship between them and the ultrasonic response, thereby achieving simultaneous identification and quantification of multi-scale impurities. Existing methods generally lack a systematic exploration of the multidimensional characteristics of ultrasonic signals, relying only on isolated information in the time or frequency domains, which cannot fully characterize the propagation and scattering behavior of weak impurity signals in complex matrix environments.

[0004] Existing technologies for detecting impurities in concentrated solutions generally suffer from the following problems: the lack of a multi-dimensional feature system encompassing the time, frequency, and wavelet domains leads to a loss of sensitive features for micro- and nano-sized impurities; the absence of an effective feature fusion mechanism to suppress ultrasonic background interference caused by the concentrated solution matrix itself results in the impurity signal being submerged; and the lack of a deep learning model with multi-scale sensing capabilities for fine segmentation of fused features makes it difficult to simultaneously distinguish between large-size aggregates and dispersed particles. These deficiencies are particularly pronounced in practical applications where impurities have a wide particle size distribution and extremely low content, resulting in large measurement deviations and poor repeatability, thus hindering the engineering application of ultrasonic testing technology in high-precision quality control systems. Therefore, a novel ultrasonic testing and data analysis method integrating multi-domain feature extraction, matrix interference suppression, and dynamic content mapping is urgently needed. Summary of the Invention

[0005] This invention provides an ultrasonic detection and data analysis method for impurity content in concentrated solutions, aiming to solve the technical problems in existing technologies that rely solely on a single ultrasonic feature, leading to difficulties in distinguishing between matrix and impurity signals, low sensitivity in identifying micro- and nano-sized impurities, susceptibility to missed or false detections, inability to adapt to scenarios with uneven impurity particle size distribution, large calculation errors for low impurity content, and inability to accurately distinguish impurity content of different particle sizes. This invention achieves high-precision identification and quantitative analysis of micro- and nano-sized impurities in concentrated solutions by constructing a multi-dimensional ultrasonic feature fusion analysis system, combined with particle size adaptive segmented modeling and a dynamic baseline correction mechanism.

[0006] This invention provides a method for ultrasonic detection and data analysis of impurity content in concentrated solutions, including: Acquire raw data of the echo signal of the concentrate under ultrasonic excitation; The original echo signal data is decomposed in the time and frequency domain to extract at least 5 types of ultrasound feature parameters; Based on the aforementioned ultrasonic feature parameters, a multidimensional feature vector is constructed and principal component analysis is performed to reduce dimensionality. The dimensionality-reduced feature set is input into a pre-trained particle size segmentation identification model to determine the particle size range to which the impurity belongs; The impurity mass concentration in each particle size range is calculated by calling the corresponding impurity-signal mapping function based on the particle size range. Perform dynamic baseline correction to weight the impurity mass concentration in each particle size range; output the weighted impurity mass concentration and total impurity content in each particle size range.

[0007] Preferably, the original echo signal data is subjected to time-frequency joint domain decomposition to extract at least 5 types of ultrasound feature parameters, including: The original data of the echo signal is decomposed in the time-frequency domain by a serial combination structure of short-time Fourier transform and continuous wavelet transform. The echo signal is then subjected to short-time Fourier transform to obtain the global spectral profile. The local transient scattering events are then finely characterized in time and frequency by applying Morlet wavelet. Based on the time-frequency joint domain decomposition results, the center frequency offset, spectral bandwidth expansion rate, attenuation coefficient gradient, scattered echo energy density, and phase coherence index are calculated as the five types of ultrasound characteristic parameters.

[0008] Preferably, the center frequency offset is defined as the absolute value of the difference between the peak frequency of the main lobe of the echo signal and the excitation center frequency; The spectral bandwidth spread rate is defined as the ratio of the -6 dB bandwidth of the echo signal to the -6 dB bandwidth of the excitation signal; The attenuation coefficient gradient is obtained by calculating the rate of change of the natural logarithm of the echo amplitude at different depth layers with the propagation distance; The energy density of the scattered echo is the square integral of the scattered echo signal within a unit time window; The phase coherence index is obtained by calculating the reciprocal of the standard deviation of the phase difference between adjacent sampling points.

[0009] Preferably, constructing a multidimensional feature vector based on the ultrasound feature parameters and performing principal component analysis to reduce dimensionality includes: The 5-dimensional feature vector composed of the 5 types of ultrasound feature parameters is subjected to Z-score normalization. Construct a covariance matrix based on normalized eigenvectors of no less than 1,000 historical samples; Solve for the eigenvalues ​​and eigenvectors of the covariance matrix, sort them in descending order of eigenvalues ​​and accumulate the variance contribution rate, and retain the top N principal components with a cumulative variance contribution rate of not less than 95% as the dimensionality-reduced feature set.

[0010] Preferably, the dimensionality-reduced feature set is input into a pre-trained particle size segmentation recognition model, including: Prepare a standard concentrate sample set covering all three particle size ranges, with each range containing no fewer than 30 samples of different concentration gradients; Ultrasonic echo signals were acquired from each sample, and the five types of ultrasonic feature parameters were extracted to construct a training feature set. The kernel function type and penalty parameter of the support vector machine are optimized by grid search. The kernel function is selected from linear kernel, polynomial kernel or radial basis function kernel, and the one with the highest cross-validation accuracy is used as the final model configuration. The trained particle size segmentation recognition model is embedded in an embedded signal processing platform to receive the dimensionality-reduced feature set and output the membership scores of the three particle size intervals. The interval corresponding to the maximum score is taken as the impurity particle size interval.

[0011] Preferably, the impurity mass concentration for each particle size range is calculated by calling the corresponding impurity-signal mapping function based on the particle size range, including: An impurity-signal mapping function is independently established for each particle size range, and its form is as follows: , The mass concentration of impurities. , , , , These are the normalized values ​​of the five types of ultrasound characteristic parameters, respectively. , , , , For regression coefficients, For bias terms; All parameters were determined by fitting the standard samples in the corresponding particle size range using the least squares method, with a goodness of fit of not less than 0.92. After removing outlier samples whose absolute residual value is greater than 3 times the standard deviation, the mapping function for the corresponding interval is called to calculate the preliminary mass concentration value.

[0012] Preferably, performing dynamic baseline correction includes: At least 100 ultrasonic echo signals were acquired in the impurity-free concentrate. The signals acquired each time were decomposed in the time and frequency domain and five types of ultrasonic feature parameters were extracted. The mean and standard deviation of each feature parameter were calculated to form a five-dimensional Gaussian distribution model as a baseline template. The squared Mahalanobis distance between the feature vector of the current echo signal and the mean vector of the baseline template is used as the residual energy. Based on the measured viscosity of the concentrate With temperature According to the formula Calculate the preset threshold , As the baseline threshold, As the reference viscosity, Reference temperature; If the residual energy is less than If so, the corrected weight is set to 1; otherwise, the corrected weight is set to 0. ; The initial mass concentration value for each particle size range is multiplied by the aforementioned correction weight to obtain the weighted corrected mass concentration value.

[0013] Preferably, the output includes the weighted adjusted impurity mass concentration and total impurity content for each particle size range, including: The total impurity content is calculated as the sum of the weighted corrected mass concentrations for each particle size range; Perform consistency check: If the relative deviation between the sum of impurity mass concentrations in any two adjacent particle size ranges and the total impurity content is greater than 5%, a secondary analysis process is triggered. The secondary analysis process includes increasing the sampling frequency to 8 kHz, re-acquiring the echo signal and repeating the aforementioned steps, up to a maximum of 3 times. If the consistency requirement is still not met, mark the result as "requires manual review" and record the original data; The final output includes the impurity mass concentration, total impurity content, detection status indicator, and environmental parameters for each particle size range.

[0014] Preferably, the ultrasonic transducer array consists of 16 piezoelectric ceramic units, which are arranged in a ring symmetrical arrangement with a center-to-center spacing of 3 mm, a working bandwidth of 400 kHz to 12 MHz, a pulse repetition frequency of 5 kHz, and a duration of 20 milliseconds for each acquisition.

[0015] Preferably, the raw echo signal data is stored in the synchronous dynamic random access memory of the embedded signal processing platform with a 16-bit sampling depth and a 40 MHz sampling rate, and the transducer array is impedance matched and calibrated before each acquisition.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention overcomes the limitation that a single feature cannot effectively characterize the scattering characteristics of micro- and nano-scale impurities by constructing a 5-dimensional ultrasonic feature system that includes center frequency offset, spectral bandwidth spread, attenuation coefficient gradient, scattered echo energy density, and phase coherence index. 2. By introducing a particle size segmentation identification model and a range-specific impurity-signal mapping function, we have achieved differentiated modeling and accurate quantification of impurities with different particle sizes, thus solving the problem of content measurement distortion in scenarios with uneven particle size distribution. 3. A dynamic baseline correction mechanism based on Mahalanobis distance is adopted, combined with real-time compensation for concentrate viscosity and temperature, which suppresses the interference of matrix fluctuations on impurity signals and improves the signal-to-noise ratio and accuracy of low impurity content detection. 4. Through consistency verification and secondary analysis processes, the reliability of the test results is further guaranteed. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram of the core principle framework of multi-dimensional ultrasonic feature fusion and particle size adaptive segmented modeling in this invention; Figure 3 This is a flowchart illustrating the logical flow of the time-frequency joint domain decomposition of echo signals and the extraction of five types of ultrasonic feature parameters in this invention. Figure 4 This is a logical flowchart of the dynamic baseline correction and impurity concentration weighted correction mechanism in this invention; Figure 5 This is a schematic diagram of the multi-level interaction relationship and data flow between the ultrasonic signal acquisition unit and each data processing functional module in this invention. Detailed Implementation

[0018] refer to Figures 1 to 5This invention provides an ultrasonic detection and data analysis method for impurity content in concentrated solutions, addressing the technical problems in existing technologies that rely solely on a single ultrasonic feature, leading to difficulties in distinguishing between matrix and impurity signals, low sensitivity in identifying micro- and nano-sized impurities, susceptibility to false negatives or missed detections, inability to adapt to scenarios with uneven impurity particle size distribution, large calculation errors for low impurity content, and inaccurate differentiation of impurity content with different particle sizes. This method constructs a multi-dimensional ultrasonic feature fusion analysis system, combined with particle size adaptive segmented modeling and a dynamic baseline correction mechanism, to achieve high-precision identification and quantitative analysis of micro- and nano-sized impurities in concentrated solutions.

[0019] The method includes the following steps: Acquire raw data of the echo signal of the concentrate under ultrasonic excitation; The original echo signal data is decomposed in the time and frequency domain to extract at least 5 types of ultrasound feature parameters; Based on the aforementioned ultrasonic feature parameters, a multidimensional feature vector is constructed and principal component analysis is performed to reduce dimensionality. The dimensionality-reduced feature set is input into a pre-trained particle size segmentation identification model to determine the particle size range to which the impurity belongs; The impurity mass concentration in each particle size range is calculated by calling the corresponding impurity-signal mapping function based on the particle size range. Perform dynamic baseline correction to weighted correct the impurity mass concentration for each particle size range; Output the weighted correction of impurity mass concentration and total impurity content for each particle size range.

[0020] Raw echo signal data of the concentrate under ultrasonic excitation was acquired. This raw echo signal data was obtained by an ultrasonic transducer array at multiple preset frequency points. These preset frequency points covered a frequency band from 500 kHz to 10 MHz, with an interval of no more than 200 kHz between adjacent frequency points. The ultrasonic transducer array consisted of 16 piezoelectric ceramic units arranged symmetrically in a ring, with a center-to-center spacing of 3 mm. The operating bandwidth was 400 kHz to 12 MHz, the pulse repetition frequency was set to 5 kHz, and each acquisition lasted for 20 milliseconds. During acquisition, the ultrasonic transducer array emitted short-pulse ultrasonic waves into the concentrate and received scattered echo signals formed by reflections from impurity particles and the matrix interface within the concentrate. All echo signals were stored digitally in the synchronous dynamic random access memory of the embedded signal processing platform, with a sampling bit depth of 16 bits and a sampling rate of 40 MHz. To ensure signal stability, the transducer array was impedance matched and calibrated before each acquisition, and the ambient temperature and concentrate viscosity were measured in real time and recorded as input parameters for subsequent dynamic baseline correction.

[0021] The raw echo signal data is decomposed in the time-frequency domain to extract at least five types of ultrasonic feature parameters. These five types of ultrasonic feature parameters include: center frequency offset, spectral bandwidth expansion rate, attenuation coefficient gradient, scattered echo energy density, and phase coherence index. The time-frequency domain decomposition employs a serial combination structure of short-time Fourier transform and continuous wavelet transform. A short-time Fourier transform is performed on the echo signal, with a window length set to four times the excitation signal period and an overlap rate of 75%, to obtain the global spectral profile. Subsequently, Morlet wavelets are applied to refine the time-frequency characterization of local transient scattering events, with wavelet scale parameters ranging from 1 to 64 and a time window length twice the signal period. This dual decomposition structure preserves both the overall spectral characteristics of the signal and captures the instantaneous details of weak scattering events.

[0022] The center frequency offset is defined as the absolute value of the difference between the peak frequency of the echo signal's main lobe and the excitation center frequency. This parameter reflects the resonant scattering effect of impurity particles on ultrasound, and is particularly sensitive to particles smaller than 500 nanometers. The spectral bandwidth spread is defined as the ratio of the echo signal's -6 dB bandwidth to the excitation signal's -6 dB bandwidth. This parameter characterizes the inelastic energy dissipation during scattering and is related to the surface roughness and internal structure of the impurity particles.

[0023] The attenuation coefficient gradient is obtained by calculating the rate of change of the natural logarithm of the echo amplitude at different depths with propagation distance. The echo signal is divided into 10 equidistant layers, each 1 mm thick. The echo amplitude at the center of each layer is calculated, and the natural logarithm is used to fit a straight line; the slope of this line is the attenuation coefficient gradient. This parameter can distinguish the energy attenuation differences caused by matrix absorption and impurity scattering.

[0024] The scattered echo energy density is the square integral of the scattered echo signal within a unit time window. The time window length is set to three times the excitation pulse width to cover the complete scattering response. The parameter is directly related to the total volume concentration of impurity particles. The phase coherence index is obtained by calculating the reciprocal of the standard deviation of the phase difference between adjacent sampling points. The phase sequence is extracted by Hilbert transform; the smaller the standard deviation, the higher the coherence, indicating that the scattering source has a stable geometric structure, suitable for distinguishing between aggregates and monodisperse particles.

[0025] A multidimensional feature vector is constructed based on the five types of ultrasound feature parameters, and principal component analysis (PCA) is performed on this multidimensional feature vector to reduce its dimensionality. The multidimensional feature vector is a 5-dimensional real vector, with each dimension corresponding to the normalized value of a type of ultrasound feature parameter. Normalization uses the Z-score method, which involves subtracting the historical sample mean and then dividing by the standard deviation. Principal component analysis is performed by solving for the eigenvalues ​​and eigenvectors of the covariance matrix. The covariance matrix is ​​obtained by statistically analyzing the feature vectors of at least 1,000 sets of samples with and without impurities. Principal components with a cumulative variance contribution rate of at least 95% are retained as the feature set after dimensionality reduction. Typically, the first three principal components are sufficient to meet the threshold requirement; therefore, the feature set after dimensionality reduction is a three-dimensional vector. This dimensionality reduction operation eliminates redundancy and noise interference between features, improving the generalization ability of subsequent classification and regression models.

[0026] The dimensionality-reduced feature set is input into a pre-trained particle size segmentation identification model. This model is a multi-classifier based on a support vector machine (SVM), with its classification boundaries divided into three intervals based on impurity particle size: less than 50 nm, 50 nm to 500 nm, and greater than 500 nm. Each interval corresponds to an independent impurity response pattern library. The model training process includes: preparing a standard concentrated solution sample set containing impurities of known particle size and concentration, covering all three particle size intervals, with each interval containing at least 30 samples of different concentration gradients; acquiring ultrasonic echo signals from each sample and extracting the five types of ultrasonic feature parameters; optimizing the kernel function type and penalty parameters of the SVM using a grid search method, with the kernel function selected from linear kernels, polynomial kernels, or radial basis function kernels, and finally choosing the configuration with the highest cross-validation accuracy as the model parameters. After training, the model is stored in the read-only memory of the embedded signal processing platform in the form of a decision function. After inputting the dimensionality-reduced feature set, the model outputs membership scores for three intervals. The interval corresponding to the maximum score is taken as the particle size interval to which the impurities in the current echo signal belong.

[0027] Based on the output of the particle size segmentation identification model, the particle size range of impurities contained in the current echo signal is determined, and the impurity-signal mapping function for the corresponding range is invoked. The impurity-signal mapping function is a nonlinear regression equation obtained by least squares fitting; its input is the dimensionality-reduced feature set, and its output is the mass concentration of impurities within that particle size range. The impurity-signal mapping function is independently established for each particle size range, and its form is: ; The mass concentration of impurities. , , , , These are the normalized values ​​of the five types of ultrasound characteristic parameters, respectively. , , , , For regression coefficients, This represents the bias term. All parameters were determined by fitting standard samples within the corresponding particle size range using the least squares method, and the goodness of fit was... Not less than 0.92. During the fitting process, outlier samples with residual absolute values ​​greater than 3 times the standard deviation are removed to ensure model robustness. After calling the mapping function, the preliminary mass concentration values ​​for the three particle size intervals are calculated and denoted as follows. .

[0028] The mass concentration and total content of unknown impurities were determined by the following experiments: For sample preparation, three groups of concentrated solution impurity samples with unknown components were selected, corresponding to three particle size ranges: Sample A (particle size < 50 nm), Sample B (particle size 50 nm - 500 nm), and Sample C (particle size > 500 nm). The particle size range of each sample was determined using a laser particle size analyzer; the specific components of the impurities were not measured. The three groups of impurities were then mixed into a pure concentrate matrix to prepare test samples with concentration gradients ranging from 0.01 mg / L to 10 mg / L, with each sample group containing 30 concentration gradients.

[0029] Ultrasonic detection and feature extraction utilize the ultrasonic transducer array of this invention to acquire echo signals from the sample under test, with a sampling rate of 40MHz and an acquisition time of 20ms. The echo signals are then subjected to joint time-frequency domain decomposition using short-time Fourier transform and Morlet wavelet transform to extract five types of ultrasonic feature parameters, which are then normalized using Z-score to obtain the normalized feature vector of sample A. The normalized feature vector of sample B The normalized feature vector of sample C .

[0030] The formula for Z-score normalization is: ; These are the original measured values ​​of the characteristic parameters. The mean of the feature parameters of no less than 1000 historical samples, The standard deviation of the feature parameters is defined for no fewer than 1000 historical samples. In this embodiment, the historical mean vectors corresponding to the five types of ultrasound feature parameters are shown. Historical standard deviation vector Taking sample A as an example, its original values ​​for the 5 feature parameters are (13.6, 12.3, 17.0, 14.1, 7.2). Substituting these values ​​into the formula, the calculation is as follows: Finally obtained The normalization calculation process for samples B and C is the same as that for sample A. Their original feature parameter values ​​are (18.9, 11.9, 20.9, 12.6, 11.8) and (22.1, 17.3, 22.2, 14.7, 14.9), respectively. Substituting these values ​​into the Z-score formula yields... , The value.

[0031] The regression coefficients of the mapping function for each particle size interval are determined by the least squares method, and the unified solution formula is as follows: In the formula, To augment the feature matrix, For the standard sample concentration vector, For the coefficient vector, It is a bias term.

[0032] ; To predict concentrations for the model, This represents the average concentration.

[0033] In this embodiment, the goodness of fit of each interval The values ​​are 0.93, 0.95, and 0.94 respectively, all of which meet the requirement of ≥0.92.

[0034] The impurity-signal mapping function for each particle size range was obtained by fitting standard samples to calculate the initial mass concentration. The goodness of fit was then determined. All values ​​are ≥0.92. The specific functions are as follows: Particle size < 50nm range: ; Particle size range of 50nm-500nm: ; Particle size > 500nm range: ; Substituting the normalized eigenvectors into the corresponding functions, the preliminary concentrations were calculated to be 0.68 mg / L (sample A), 3.25 mg / L (sample B), and 8.72 mg / L (sample C).

[0035] Substituting sample A (particle size < 50 nm) into the corresponding function: ; Substituting sample B (particle size 50nm-500nm) into the corresponding function: ; Substituting sample C (particle size > 500 nm) into the corresponding function: .

[0036] Dynamic baseline correction was performed by acquiring 100 sets of baseline signals from the impurity-free concentrate, constructing a 5-dimensional Gaussian distribution baseline template, and calculating the mean vector. The covariance matrix Σ is a diagonal matrix, and the diagonal elements are the variances of each characteristic parameter.

[0037] The mean vector μ is the arithmetic mean of the feature vectors of 100 baseline samples, calculated using the following formula: ; This represents the baseline sample size. Corresponding to 5 types of ultrasound characteristic parameters, For the first The first group of baseline samples Normalized values ​​of class feature parameters. In this embodiment, the average values ​​of the five class feature parameters of 100 baseline samples are calculated respectively to obtain the mean vector. Covariance matrix It is a diagonal matrix, with diagonal elements representing the variances of the characteristic parameters. In this embodiment Inverse matrix .

[0038] The formula for calculating the squared Mahalanobis distance is: In the formula, For sample feature vectors, This represents the baseline mean vector. The calculation process for the squared Mahalanobis distance of the three samples in this embodiment is as follows: Sample A: ; ; Sample B: ; ; Sample C: ; .

[0039] Measure the current viscosity of the concentrate Temperature (°C), reference threshold, reference viscosity, and reference temperature are respectively Substitute into the formula to calculate the threshold: Calculate the squared Mahalanobis distance between the feature vector of sample A and the baseline template. less than the threshold Adjust weights The corrected concentration was 0.68 mg / L; for sample B... greater than the threshold Adjust weights The corrected concentration was 2.99 mg / L; for sample C... If the value is less than the threshold T, adjust the weight. The corrected concentration is 8.72 mg / L.

[0040] The total impurity content is calculated as the sum of the corrected concentrations for each particle size range, which is 12.39 mg / L.

[0041] After determining the impurity mass concentration for each particle size range, a dynamic baseline correction operation is performed. This dynamic baseline correction operation establishes a baseline template based on historical echo signals of the concentrate matrix in an impurity-free state. The baseline template is established as follows: at least 100 ultrasonic echo signal acquisitions are performed in the impurity-free concentrate. For each acquired signal, the aforementioned time-frequency joint domain decomposition is performed, and five types of ultrasonic characteristic parameters are extracted. The mean and standard deviation of each characteristic parameter are calculated to form a 5-dimensional Gaussian distribution model, which serves as the baseline template. The residual energy is defined as the squared Mahalanobis distance between the current echo signal characteristic vector and the mean vector of the baseline template. The formula is dynamically adjusted based on the measured values ​​of the concentrate viscosity and temperature, and the calculation is as follows: ; The baseline threshold is set at 2.8. The measured viscosity of the concentrate is given in millipascals per second. 0 is the reference viscosity, with a value of 1.5; The current measured temperature of the concentrate is in degrees Celsius. The reference temperature is 25 degrees Celsius. If the residual energy... Less than If the matrix fluctuation is within the normal range, the correction weight is set to 1; if Greater than or equal to Then adjust the weights. The impurity mass concentrations for each particle size range are updated after weighted correction. , This is the weighting coefficient. The mechanism suppresses spurious signals caused by changes in matrix composition, bubble incorporation, or temperature drift, improving the reliability of low-concentration impurity detection.

[0042] Output the weighted adjusted impurity mass concentration and total impurity content for each particle size range. Total impurity content Before output, a consistency check is performed: if the relative deviation between the sum of impurity mass concentrations in any two adjacent particle size ranges and the total impurity content is greater than 5%, a secondary analysis process is triggered. The formula for calculating the relative deviation is... , For the first The updated impurity mass concentration after weighted correction. For the first The weighted and corrected impurity mass concentration is updated. When the trigger condition is met, the system automatically increases the sampling frequency to 8 kHz, reacquires the echo signal, and repeats steps S1 to S7 until the consistency requirement is met. The secondary analysis can be performed a maximum of 3 times. If the requirement is still not met, the result is marked as "requires manual verification," and the original data is recorded for subsequent diagnosis. The final output includes: impurity mass concentrations smaller than 50 nm, impurity mass concentrations from 50 nm to 500 nm, impurity mass concentrations larger than 500 nm, total impurity content, detection status indicator, and environmental parameters (temperature, viscosity).

[0043] The method operates on an embedded signal processing platform equipped with a dual-core digital signal processor with a clock frequency of 800 MHz and 512 megabytes of synchronous dynamic random access memory. The ultrasonic signal acquisition unit, multi-dimensional feature extraction unit, feature dimensionality reduction unit, particle size segmentation identification unit, concentration mapping calculation unit, dynamic baseline correction unit, and result output unit are connected via an internal data bus. Each unit runs on an independent thread, and a circular buffer is used to synchronize the data flow. The entire analysis process, from signal acquisition to result output, has a latency of no more than 300 milliseconds, meeting the requirements for online real-time detection.

[0044] In scenarios with highly uneven impurity particle size distribution, such as the simultaneous presence of 10-nanometer aggregates and 800-nanometer crystal particles, this method can still accurately separate the contributions of each region, avoiding the overestimation or underestimation of concentration caused by signal superposition in traditional single-feature methods. The dynamic baseline correction mechanism enables the system to maintain stable detection accuracy even when the concentrate batch is changed or the ambient temperature fluctuates by 10 degrees Celsius, which is superior to the fixed threshold correction scheme.

[0045] In summary, this embodiment achieves high sensitivity, high selectivity, and high precision detection of micro- and nano-scale impurities in concentrates through a three-in-one technical architecture of multi-dimensional ultrasonic feature fusion, particle size adaptive segmented modeling, and dynamic baseline correction, comprehensively solving the applicability bottleneck of existing technologies in complex industrial scenarios.

[0046] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0047] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for ultrasonic detection and data analysis of impurity content in concentrated liquid, characterized in that, include: Acquire raw data of the echo signal of the concentrate under ultrasonic excitation; The original echo signal data is decomposed in the time and frequency domain to extract at least 5 types of ultrasound feature parameters; Based on the aforementioned ultrasonic feature parameters, a multidimensional feature vector is constructed and principal component analysis is performed to reduce dimensionality. The dimensionality-reduced feature set is input into a pre-trained particle size segmentation identification model to determine the particle size range to which the impurity belongs; The impurity mass concentration in each particle size range is calculated by calling the corresponding impurity-signal mapping function based on the particle size range. Perform dynamic baseline correction to weight the impurity mass concentration in each particle size range; output the weighted impurity mass concentration and total impurity content in each particle size range.

2. The method for ultrasonic detection and data analysis of impurity content in concentrated liquid according to claim 1, characterized in that, The raw echo signal data is decomposed in the time and frequency domains to extract at least five types of ultrasound feature parameters, including: The original data of the echo signal is decomposed in the time-frequency domain by a serial combination structure of short-time Fourier transform and continuous wavelet transform. The echo signal is then subjected to short-time Fourier transform to obtain the global spectral profile. The local transient scattering events are then finely characterized in time and frequency by applying Morlet wavelet. Based on the time-frequency joint domain decomposition results, the center frequency offset, spectral bandwidth expansion rate, attenuation coefficient gradient, scattered echo energy density, and phase coherence index are calculated as the five types of ultrasound characteristic parameters.

3. The method for ultrasonic detection and data analysis of impurity content in concentrated liquid according to claim 2, characterized in that, The center frequency offset is defined as the absolute value of the difference between the peak frequency of the main lobe of the echo signal and the excitation center frequency. The spectral bandwidth spread rate is defined as the ratio of the -6 dB bandwidth of the echo signal to the -6 dB bandwidth of the excitation signal; The attenuation coefficient gradient is obtained by calculating the rate of change of the natural logarithm of the echo amplitude at different depth layers with the propagation distance; The energy density of the scattered echo is the square integral of the scattered echo signal within a unit time window; The phase coherence index is obtained by calculating the reciprocal of the standard deviation of the phase difference between adjacent sampling points.

4. The method for ultrasonic detection and data analysis of impurity content in concentrated liquid according to claim 3, characterized in that, Based on the aforementioned ultrasound feature parameters, a multidimensional feature vector is constructed and principal component analysis is performed to reduce dimensionality, including: The 5-dimensional feature vector composed of the 5 types of ultrasound feature parameters is subjected to Z-score normalization. Construct a covariance matrix based on normalized eigenvectors of no less than 1,000 historical samples; Solve for the eigenvalues ​​and eigenvectors of the covariance matrix, sort them in descending order of eigenvalues ​​and accumulate the variance contribution rate, and retain the top N principal components with a cumulative variance contribution rate of not less than 95% as the dimensionality-reduced feature set.

5. The method for ultrasonic detection and data analysis of impurity content in concentrated liquid according to claim 4, characterized in that, The dimensionality-reduced feature set is input into a pre-trained particle size segmentation recognition model, including: Prepare a standard concentrate sample set covering all three particle size ranges, with each range containing no fewer than 30 samples of different concentration gradients; Ultrasonic echo signals were acquired from each sample, and the five types of ultrasonic feature parameters were extracted to construct a training feature set. The kernel function type and penalty parameter of the support vector machine are optimized by grid search. The kernel function is selected from linear kernel, polynomial kernel or radial basis function kernel, and the one with the highest cross-validation accuracy is used as the final model configuration. The trained particle size segmentation recognition model is embedded in an embedded signal processing platform to receive the dimensionality-reduced feature set and output the membership scores of the three particle size intervals. The interval corresponding to the maximum score is taken as the impurity particle size interval.

6. The method for ultrasonic detection and data analysis of impurity content in concentrated liquid according to claim 5, characterized in that, Based on the particle size range, the corresponding impurity-signal mapping function is called to calculate the impurity mass concentration for each particle size range, including: An impurity-signal mapping function is independently established for each particle size range, and its form is as follows: , The mass concentration of impurities. , , , , These are the normalized values ​​of the five types of ultrasound characteristic parameters, respectively. , , , , For regression coefficients, For bias terms; All parameters were determined by fitting the standard samples in the corresponding particle size range using the least squares method, with a goodness of fit of not less than 0.

92. After removing outlier samples whose absolute residual value is greater than 3 times the standard deviation, the mapping function for the corresponding interval is called to calculate the preliminary mass concentration value.

7. The method for ultrasonic detection and data analysis of impurity content in concentrated liquid according to claim 6, characterized in that, Perform dynamic baseline correction operations, including: At least 100 ultrasonic echo signals were acquired in the impurity-free concentrate. The signals acquired each time were decomposed in the time and frequency domain and five types of ultrasonic feature parameters were extracted. The mean and standard deviation of each feature parameter were calculated to form a five-dimensional Gaussian distribution model as a baseline template. The squared Mahalanobis distance between the feature vector of the current echo signal and the mean vector of the baseline template is used as the residual energy. Based on the measured viscosity of the concentrate With temperature According to the formula Calculate the preset threshold , As the baseline threshold, As the reference viscosity, Reference temperature; If the residual energy is less than If so, the corrected weight is set to 1; otherwise, the corrected weight is set to 0. ; The initial mass concentration value for each particle size range is multiplied by the aforementioned correction weight to obtain the weighted corrected mass concentration value.

8. The method for ultrasonic detection and data analysis of impurity content in concentrated liquid according to claim 7, characterized in that, Output the weighted adjusted impurity mass concentration and total impurity content for each particle size range, including: The total impurity content is calculated as the sum of the weighted corrected mass concentrations for each particle size range; Perform consistency check: If the relative deviation between the sum of the impurity mass concentrations of any two adjacent particle size ranges and the total impurity content is greater than 5%, a secondary analysis process is triggered. The secondary analysis process includes increasing the sampling frequency to 8 kHz, re-acquiring the echo signal and repeating the aforementioned steps, up to a maximum of 3 times. If the consistency requirement is still not met, mark the result as "requires manual review" and record the original data; The final output includes the impurity mass concentration, total impurity content, detection status indicator, and environmental parameters for each particle size range.

9. The method for ultrasonic detection and data analysis of impurity content in concentrated liquid according to claim 8, characterized in that, The raw echo signal data is acquired by an ultrasonic transducer array at multiple preset frequency points. The ultrasonic transducer array consists of 16 piezoelectric ceramic units, which are arranged in a ring symmetrical arrangement with a center spacing of 3 mm. The operating bandwidth is 400 kHz to 12 MHz, the pulse repetition frequency is set to 5 kHz, and the duration of each acquisition is 20 milliseconds.

10. The method for ultrasonic detection and data analysis of impurity content in concentrated liquid according to claim 9, characterized in that, The raw echo signal data is stored in the synchronous dynamic random access memory of the embedded signal processing platform with a 16-bit sampling depth and a 40 MHz sampling rate. The transducer array is impedance matched and calibrated before each acquisition.

Images