Titanium alloy wire performance detection method, device and system

Abstract

The application relates to the technical field of titanium alloy detection, and discloses a titanium alloy wire performance detection method, equipment and system. The method collects real-time force value displacement data of a tensile test and echo signal data of ultrasonic detection, obtains smooth force value displacement data and frequency spectrum characteristic parameters after processing, fuses yield strength and tensile strength with peak frequency and amplitude attenuation rate in a characteristic layer, and constructs a first multi-dimensional characteristic matrix; the method inputs a pre-trained deep confidence network model, predicts mechanical properties such as elongation and hardness value, compares with a preset threshold value, marks unqualified items, and generates a detection report. The method improves the comprehensiveness and accuracy of detection through multi-source data fusion and advanced algorithm processing, realizes efficient evaluation of titanium alloy wire performance, and is suitable for strict detection requirements of titanium alloy wire quality in the fields of aerospace, medical devices and the like.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of titanium alloy detection, in particular to a titanium alloy wire performance detection method, device and system. BACKGROUND

[0002] Titanium alloy wires are widely used in aerospace, medical devices, high-end manufacturing and other fields due to their high strength, corrosion resistance and lightweight characteristics. These application scenarios have high requirements for the mechanical properties of titanium alloy wires, and the stability and reliability of the performance directly relate to the quality and safety of the end products. Therefore, accurate and efficient detection of the performance of titanium alloy wires is an important link in the production and application process.

[0003] Currently, the performance detection of titanium alloy wires mainly relies on traditional physical test methods, such as tensile test, hardness test, etc. Tensile test measures the yield strength, tensile strength and elongation of the wire before it breaks by applying axial tension to the wire, which is a common method for evaluating the mechanical properties of materials. However, this method has obvious limitations: tensile test is a destructive test, and the wire after detection cannot be used again, resulting in waste of raw materials, especially for high-value titanium alloy materials, which will significantly increase production costs; the test process requires manual operation and data recording, which not only takes a long time, but also is easily affected by the skill level of the operator and environmental factors, making it difficult to ensure the consistency and accuracy of the detection results.

[0004] In addition to tensile test, ultrasonic detection technology is also used for quality evaluation of titanium alloy wires, mainly for detecting internal defects of the material such as cracks, pores, etc. Traditional ultrasonic detection focuses on the time domain characteristics of the signal, and analyzes the amplitude and propagation time of the echo signal to determine the presence or absence of defects. However, this method lacks correlation analysis of material mechanical properties and cannot directly reflect key mechanical indicators such as yield strength and hardness of the wire, limiting the application range of the detection results.

[0005] With the development of industrial intelligence, some detection methods combining sensor technology and data processing algorithms have emerged. However, in existing technologies, most methods rely on only a single type of detection data for performance evaluation, such as using only tensile test data or relying only on ultrasonic detection signals, which cannot fully reflect the comprehensive performance of titanium alloy wires. At the same time, the noise suppression effect in the data processing process is not good, and the feature parameter extraction is not comprehensive enough, resulting in low accuracy of performance prediction. In addition, traditional machine learning models often have insufficient generalization ability when dealing with multi-dimensional and nonlinear feature data, which cannot meet the demand of high-precision detection.

[0006] In actual production, due to the complex production process of titanium alloy wire, involving smelting, rolling, heat treatment and other links, the fluctuation of parameters in any link may cause the difference of material performance. The existing detection method is difficult to realize the rapid and comprehensive evaluation of the performance of the wire, and often appears the situation of missed detection or misjudgment, which not only affects the production efficiency, but also may lead to unqualified products flowing into the market, causing safety hazards. Therefore, developing a titanium alloy wire performance detection method which can integrate various detection data, improve detection accuracy and efficiency, and realize non-destructive or semi-destructive detection, has become a problem that needs to be solved by the current technical personnel in the field. SUMMARY

[0007] The purpose of the present application is to provide a titanium alloy wire performance detection method, device and system to solve the problems raised in the background art.

[0008] To achieve the above-mentioned purpose, the present application provides a titanium alloy wire performance detection method, which comprises:

[0009] Collecting real-time force value displacement data of titanium alloy wire to be detected in the tensile test process and echo signal data returned by the ultrasonic detection device, and transmitting the real-time force value displacement data and the echo signal data to a data processing unit;

[0010] The data processing unit adopts an adaptive median filtering algorithm to process the real-time force value displacement data for denoising, and obtains smooth force value displacement data, while performing fast Fourier transform on the echo signal data to extract the amplitude spectrum and phase spectrum of the signal as frequency spectrum characteristic parameters;

[0011] Fusing the yield strength and tensile strength parameters in the smooth force value displacement data with the peak frequency and amplitude decay rate parameters in the frequency spectrum characteristic parameters to construct a first multi-dimensional feature matrix;

[0012] Inputting the first multi-dimensional feature matrix into a pre-trained deep belief network model, and performing prediction processing on the elongation and hardness value of the titanium alloy wire by the deep belief network model to obtain a mechanical property prediction result;

[0013] Comparing the mechanical property prediction result with a preset performance threshold range, marking as unqualified when the mechanical property prediction result exceeds the preset performance threshold range, and generating a titanium alloy wire performance detection report containing unqualified items and corresponding parameters.

[0014] Preferably, the first multi-dimensional feature matrix is input into a pre-trained deep belief network model, and the elongation and hardness value of the titanium alloy wire are predicted by the deep belief network model to obtain a mechanical property prediction result, which comprises:

[0015] obtaining a multi-dimensional feature matrix corresponding to a historical titanium alloy wire sample and measured mechanical property data from a database, and dividing the historical titanium alloy wire sample into a training set and a validation set;

[0016] unsupervised pre-training an initial deep belief network model using the training set, extracting features layer by layer through stacking a restricted Boltzmann machine, and fine-tuning parameters of the initial deep belief network model using a back propagation algorithm;

[0017] evaluating performance of the fine-tuned deep belief network model using the validation set, calculating a prediction error, stopping training when the prediction error is less than a preset error threshold, and obtaining the pre-trained deep belief network model;

[0018] inputting the first multi-dimensional feature matrix into the pre-trained deep belief network model, and outputting an elongation prediction value and a hardness prediction value of the titanium alloy wire as the mechanical property prediction result.

[0019] Preferably, the unsupervised pre-training of the initial deep belief network model using the training set, the extraction of features layer by layer through stacking the restricted Boltzmann machine, and the fine-tuning of the parameters of the initial deep belief network model using the back propagation algorithm, comprises:

[0020] initializing the number of neurons of the input layer, the hidden layer and the output layer of the initial deep belief network model, setting the learning rate and the number of iterations of the restricted Boltzmann machine;

[0021] inputting the multi-dimensional feature matrix in the training set into the first layer of the restricted Boltzmann machine, training through the contrastive divergence algorithm, and obtaining the first layer feature mapping;

[0022] training the first layer feature mapping as the input of the second layer of the restricted Boltzmann machine, and sequentially stacking to a preset number of layers to complete the unsupervised pre-training process;

[0023] using the measured mechanical property data of the training set as labels, calculating the mean square error of the prediction value and the labels through the back propagation algorithm, adjusting the connection weights and bias values between the neurons of each layer according to the mean square error, and completing the parameter fine-tuning.

[0024] Preferably, the use of the measured mechanical property data of the training set as labels, the calculation of the mean square error of the prediction value and the labels through the back propagation algorithm, the adjustment of the connection weights and bias values between the neurons of each layer according to the mean square error, and the completion of the parameter fine-tuning, comprises:

[0025] inputting the multi-dimensional feature matrix of the training set into the deep belief network model that has completed the unsupervised pre-training, and obtaining the predicted mechanical property data of each sample.

[0026] calculating the mean square error between the predicted mechanical property data and the measured mechanical property data, and taking the mean square error as a loss function;

[0027] deriving the loss function by using the gradient descent method to obtain the gradient direction of the connection weight and the bias value of each layer;

[0028] updating the connection weight and the bias value according to the gradient direction with a preset learning rate, and repeating iteration until the loss function converges to a preset minimum value range.

[0029] Preferably, the yield strength and tensile strength parameters in the smoothed force value displacement data are fused with the peak frequency and amplitude attenuation rate parameters in the spectral feature parameters to construct a first multi-dimensional feature matrix, including:

[0030] normalizing the yield strength and tensile strength parameters in the smoothed force value displacement data to obtain standardized mechanical parameters;

[0031] standardizing the peak frequency and amplitude attenuation rate parameters in the spectral feature parameters to obtain standardized spectral parameters;

[0032] concatenating the standardized mechanical parameters and the standardized spectral parameters by column to form an initial feature matrix;

[0033] using an independent component analysis algorithm to reduce the dimension of the initial feature matrix to remove redundant features and obtain a first multi-dimensional feature matrix.

[0034] Preferably, the independent component analysis algorithm is used to reduce the dimension of the initial feature matrix to remove redundant features and obtain a first multi-dimensional feature matrix, including:

[0035] centering the initial feature matrix to make the mean value of each feature zero;

[0036] calculating the covariance matrix of the centered initial feature matrix, and performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors;

[0037] sorting the eigenvalues according to their sizes, selecting the first k eigenvectors whose cumulative contribution rate reaches a preset threshold, and constructing a projection matrix;

[0038] multiplying the centered initial feature matrix by the projection matrix to obtain a reduced multi-dimensional feature matrix.

[0039] Preferably, as spectral feature parameters, including:

[0040] The echo signal data is truncated to remove noise segments at the beginning and end of the signal and obtain an effective signal segment;

[0041] A Hanning window is added to the effective signal segment to reduce spectral leakage and obtain a windowed signal;

[0042] Fast Fourier transform is performed on the windowed signal to convert the time-domain signal into a frequency-domain signal;

[0043] The frequency corresponding to the maximum peak value of the amplitude spectrum is extracted from the frequency-domain signal as a peak frequency parameter, and the frequency range at which the amplitude decays to half the peak value is calculated as an amplitude decay rate parameter.

[0044] Preferably, the real-time force value displacement data of the titanium alloy wire during the tensile test and the echo signal data returned by the ultrasonic detection device are collected, including:

[0045] The real-time force value data is obtained by a tensile-compressive force sensor, and the displacement data is collected by a laser range finder.

[0046] The ultrasonic detection device is controlled to emit ultrasonic pulses at a preset frequency to different radial sections of the titanium alloy wire, and receive echo signal data returned by each section, wherein the preset frequency is dynamically adjusted according to the diameter specification of the titanium alloy wire.

[0047] Timestamp labels are added to the real-time force value data, displacement data and echo signal data to establish a time correlation index of multiple sources of data.

[0048] Preferably, the present application also includes a titanium alloy wire performance detection system for implementing the titanium alloy wire performance detection method as described above, the system comprising:

[0049] The acquisition module is used to collect the real-time force value displacement data of the titanium alloy wire during the tensile test and the echo signal data returned by the ultrasonic detection device, and transmit the real-time force value displacement data and the echo signal data to the data processing unit.

[0050] The preprocessing module is used to perform denoising processing on the real-time force value displacement data by the data processing unit to obtain smoothed force value displacement data, and simultaneously perform fast Fourier transform on the echo signal data to extract spectral feature parameters.

[0051] The fusion module is used to fuse the yield strength and tensile strength parameters in the smoothed force value displacement data with the peak frequency and amplitude decay rate parameters in the spectral feature parameters in a feature layer to construct a first multi-dimensional feature matrix.

[0052] The prediction module is configured to input the first multi-dimensional feature matrix into a pre-trained deep belief network model, and perform prediction processing on the elongation and hardness value of the titanium alloy wire through the deep belief network model to obtain a mechanical property prediction result.

[0053] The report generation module is configured to compare the mechanical property prediction result with a preset performance threshold range, mark the titanium alloy wire as unqualified when the mechanical property prediction result exceeds the preset performance threshold range, and generate a titanium alloy wire performance detection report containing unqualified items and corresponding parameters.

[0054] Preferably, the present application further comprises a computing device, which comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to realize the titanium alloy wire performance detection method as described above.

[0055] Compared with the prior art, the present application has the following beneficial effects:

[0056] By integrating the real-time force value displacement data of the tensile test and the echo signal data of the ultrasonic detection, effective fusion of multi-source information is realized, and the limitations of traditional single detection method are overcome. Traditional detection methods often rely only on one of the data of tensile test or ultrasonic detection, resulting in insufficient evaluation of material performance. However, the present method fuses the two kinds of data at the feature layer to construct a first multi-dimensional feature matrix, which can more comprehensively reflect the physical characteristics and internal structure information of the titanium alloy wire, thereby providing a richer basis for subsequent performance prediction.

[0057] In the data processing link, the self-adaptive median filtering algorithm is used to denoise the real-time force value displacement data. Compared with traditional filtering methods, the self-adaptive median filtering algorithm can better preserve the key features in the data while effectively eliminating noise interference, obtaining smoother and more accurate force value displacement data, and providing a reliable foundation for the extraction of important parameters such as yield strength and tensile strength. Fast Fourier transform is performed on the echo signal data to extract amplitude spectrum and phase spectrum as frequency spectrum feature parameters, which can deeply excavate the material internal structure information contained in the ultrasonic signal, such as grain size and defect distribution. These information are closely related to the mechanical properties of the material, further improving the representativeness of the feature parameters.

[0058] The key mechanical parameters in the smoothed force value displacement data are fused with the frequency spectrum feature parameters to construct a first multi-dimensional feature matrix, realizing the complementary advantages of different types of feature parameters. Yield strength, tensile strength and other parameters directly reflect the mechanical bearing capacity of the material, while peak frequency, amplitude decay rate and other frequency spectrum parameters can reflect the microstructure characteristics of the material. The combination of the two makes the feature matrix contain more comprehensive information, providing a more solid data foundation for subsequent performance prediction.

[0059] The pre-trained deep belief network model is introduced for mechanical property prediction, which fully utilizes the advantages of the model in handling nonlinear and high-dimensional data. Through multiple layers of nonlinear transformation, the deep belief network can automatically learn the complex correlation in the feature matrix, and has higher prediction accuracy and stronger generalization ability compared with traditional machine learning models. By predicting the elongation and hardness values, key mechanical property parameters can be quickly obtained without destructive testing, reducing material waste and improving detection efficiency.

[0060] During the detection process, by comparing the mechanical property prediction results with the preset threshold range, it can quickly judge whether the titanium alloy wire is qualified or not, and generate a detection report containing specific unqualified items and corresponding parameters, providing clear and intuitive reference for quality control in the production process. This method not only improves the traceability of the detection results, but also helps production personnel to timely find possible problems in the production process, facilitates the adjustment and optimization of process parameters, and reduces the production of unqualified products.

[0061] The entire detection process realizes the automation of data acquisition, processing, analysis and report generation, reduces manual intervention, reduces the influence of human factors on the detection results, and improves the consistency and reliability of the detection. At the same time, this method is suitable for the detection of titanium alloy wires of different specifications and different production batches, has strong universality and practicality, and can meet the quality detection needs in large-scale industrial production, helping to improve the overall quality level of titanium alloy wire production. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 The working principle diagram of the titanium alloy wire performance detection method described in the present application;

[0063] Figure 2 The flowchart of deep belief network model training and prediction;

[0064] Figure 3 The flowchart of deep belief network unsupervised pre-training;

[0065] Figure 4 The flowchart of the first multi-dimensional feature matrix construction;

[0066] Figure 5 The flowchart of echo signal spectrum feature extraction. DETAILED DESCRIPTION

[0067] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0068] Please refer to Figure 1 The present application provides a titanium alloy wire performance detection method, device and system, the method comprises:

[0069] Step 1: collecting real-time force value displacement data of titanium alloy wire to be detected in the tensile test process and echo signal data returned by the ultrasonic detection device, and transmitting the real-time force value displacement data and the echo signal data to a data processing unit.

[0070] Step 2: the data processing unit adopts an adaptive median filtering algorithm to process the real-time force value displacement data, obtains smooth force value displacement data, and simultaneously executes fast Fourier transform on the echo signal data, extracts the amplitude spectrum and phase spectrum of the signal as the frequency spectrum characteristic parameters.

[0071] Step 3: the yield strength and tensile strength parameters in the smooth force value displacement data are fused with the peak frequency and amplitude decay rate parameters in the frequency spectrum characteristic parameters to construct a first multi-dimensional feature matrix.

[0072] Step 4: the first multi-dimensional feature matrix is input into a pre-trained deep belief network model, and the extension rate and hardness value of the titanium alloy wire are predicted by the deep belief network model, and a mechanical property prediction result is obtained.

[0073] Step 5: comparing the mechanical property prediction result with a preset performance threshold range, marking as unqualified when the mechanical property prediction result exceeds the preset performance threshold range, and generating a titanium alloy wire performance detection report containing unqualified items and corresponding parameters.

[0074] Embodiment 1: please refer to Figure 2 The first multi-dimensional feature matrix is input into a pre-trained deep belief network model, and the extension rate and hardness value of the titanium alloy wire to be detected are predicted by the deep belief network model, and a mechanical property prediction result is obtained, and the specific process is as follows:

[0075] The multi-dimensional feature matrix and the measured mechanical property data corresponding to the historical titanium alloy wire sample are obtained from the database, and the historical titanium alloy wire sample is divided into a training set and a validation set. During the division process, the number of samples and the feature distribution should be considered comprehensively, and an appropriate division ratio should be selected. After the division is completed, it is necessary to check whether the distribution of the training set and the validation set on each feature parameter is consistent to avoid deviation in model training due to differences in data distribution.

[0076] The initial deep belief network model is pre-trained unsupervisedly using the training set, and features are extracted layer by layer by stacking restricted Boltzmann machines. As the basic unit of deep belief network, each layer of restricted Boltzmann machine is composed of visible layer and hidden layer, and there is no connection between neurons in the layer, but full connection between neurons in different layers. When training the first layer of restricted Boltzmann machine, the multi-dimensional feature matrix in the training set is input as the visible layer, and the connection weight between the visible layer and the hidden layer is adjusted through learning, so that the hidden layer can capture the basic features in the input data. After the first layer is trained, the output of the hidden layer of this layer is input as the visible layer of the second layer of restricted Boltzmann machine, and the same training process is performed to extract more abstract features. Restricted Boltzmann machines are stacked layer by layer in this way until the preset number of layers is reached, and the unsupervised pre-training is completed. After unsupervised pre-training, the parameters of the initial deep belief network model are fine-tuned using the back propagation algorithm. The back propagation algorithm calculates the error between the model output and the actual value, and transmits the error from the output layer to the input layer layer by layer. According to the error size, the connection weight and bias value between neurons in each layer are adjusted, so that the prediction result of the model is closer to the actual value.

[0077] The performance of the fine-tuned deep belief network model is evaluated using the validation set, and the prediction error is calculated. The calculation of prediction error needs to select appropriate evaluation indicators. By inputting the multi-dimensional feature matrix in the validation set into the model, the predicted mechanical property data is obtained, and then compared with the measured mechanical property data in the validation set, the error value is calculated. During the evaluation process, the change of error needs to be monitored continuously. When the prediction error is less than the preset error threshold, the training is stopped, and the pre-trained deep belief network model is obtained. If the prediction error does not reach the preset error threshold, the parameters of the model need to be adjusted, such as learning rate, iteration times, etc., and the training and evaluation are performed again until the error meets the requirements.

[0078] The first multi-dimensional feature matrix is input into the pre-trained deep belief network model, and the model performs multi-layer processing on the input feature matrix. It can be understood that the first multi-dimensional feature matrix is the multi-dimensional feature matrix corresponding to the titanium alloy wire to be detected. After the input layer receives the first multi-dimensional feature matrix, the data is transmitted to the first hidden layer. The hidden layer calculates the input data through a pre-set activation function, extracts features, and transmits them to the next hidden layer. Each hidden layer processes the data in turn, gradually refining the key features related to the elongation and hardness value of the titanium alloy wire. Finally, the output layer processes the data transmitted by the last hidden layer, and outputs the elongation prediction value and hardness prediction value of the titanium alloy wire to be detected as the mechanical property prediction result. During the entire prediction process, the model keeps the parameters determined during training unchanged to ensure the stability and consistency of the prediction result.

[0079] During the training and prediction of the model, attention should be paid to the integrity and accuracy of the data. For historical sample data in the database, strict screening and verification should be performed to remove data with missing or errors to avoid adverse effects on model training. At the same time, when dividing the training set and the validation set, a random division method should be used to avoid data bias caused by human factors. During the unsupervised pre-training and parameter fine-tuning stage, the number of iterations for training should be reasonably set to prevent overfitting or underfitting of the model. Overfitting will cause the model to perform well on the training set but poorly on the validation set and new data; underfitting will cause the model to fail to fully learn the features in the data, resulting in inaccurate prediction results. In addition, during the performance evaluation stage, the amount of data in the validation set should be sufficient to accurately reflect the generalization ability of the model. If the amount of data in the validation set is too small, the evaluation result may fluctuate greatly, making it difficult to reliably judge the performance of the model. Through the above process, a stable and reliable deep belief network model can be constructed for predicting the elongation and hardness value of titanium alloy wire, providing effective support for the performance detection of titanium alloy wire.

[0080] Example 2: Referring to Figure 3 , the initial deep belief network model is pre-trained using the training set, features are extracted layer by layer through stacking restricted Boltzmann machines, and the parameters of the initial deep belief network model are fine-tuned using the backpropagation algorithm, as follows:

[0081] The number of neurons of the input layer, hidden layer and output layer of the initial deep belief network model is initialized. The number of neurons of the input layer needs to match the dimension of the first multi-dimensional feature matrix, that is, the feature matrix contains how many feature parameters, and the input layer is set with the corresponding number of neurons. The setting of the hidden layer needs to consider the complexity of the features, and multiple hidden layers can be set, and the number of neurons of each layer can be adjusted according to the number of neurons of the previous layer, and is usually gradually reduced or kept stable with the increase of the number of layers. The number of neurons of the output layer needs to be consistent with the number of mechanical property parameters to be predicted. Since two parameters of elongation and hardness need to be predicted, two neurons are set in the output layer. At the same time, the learning rate and the number of iterations of the restricted Boltzmann machine are determined. The value of the learning rate needs to be considered comprehensively considering the training speed and stability, and the number of iterations is set according to the convergence of the model in the training process.

[0082] The multi-dimensional feature matrix in the training set is input into the first layer of the restricted Boltzmann machine, and is trained by the contrastive divergence algorithm. The running process of the contrastive divergence algorithm is as follows: first, the data in the multi-dimensional feature matrix is taken as the input of the visible layer, the activation probability of the hidden layer neuron is calculated according to the state of the visible layer neuron, and then the state of the hidden layer is determined. Then, the state of the visible layer is reconstructed according to the state of the hidden layer to obtain the reconstructed data. By comparing the difference between the original input data and the reconstructed data, the connection weight and bias value between the visible layer and the hidden layer are adjusted. Repeat the above process until the difference between the reconstructed data and the original input data reaches a stable state, at which time the training of the first layer of the restricted Boltzmann machine is completed, and the feature mapping of the first layer is obtained. The first layer feature mapping is taken as the input of the second layer of the restricted Boltzmann machine, and the same contrastive divergence algorithm is used for training to adjust the connection weight and bias value of this layer, and the feature mapping of the second layer is obtained. In the same way, the restricted Boltzmann machines are stacked layer by layer until the training of the preset number of layers is completed, and the unsupervised pre-training process is completed.

[0083] The measured mechanical property data of the training set is taken as a label, the mean square error of the predicted value and the label is calculated by the back propagation algorithm, the connection weight and the bias value between the neurons of each layer are adjusted according to the mean square error, and the parameter fine-tuning is completed. The specific operation is as follows: the multi-dimensional feature matrix of the training set is input into the deep belief network model which has completed unsupervised pre-training, and the model outputs the predicted mechanical property data of each sample through the calculation and processing of each layer of neurons. The predicted mechanical property data is compared with the corresponding measured mechanical property data, and the mean square error between them is calculated as the loss function for measuring the prediction effect of the model. The gradient descent method is used to derive the loss function, and the gradient change direction of each layer connection weight and bias value is obtained. According to the gradient direction, the connection weight and the bias value are updated according to the preset learning rate. After updating, the multi-dimensional feature matrix of the training set is input into the model again, and the new loss function value is calculated. Repeat the above process of calculating the loss function, derivation and parameter updating until the value of the loss function converges to the preset minimum value range. During the parameter fine-tuning process, the trend of the loss function needs to be paid attention to. If the loss function value no longer decreases obviously or fluctuates, the learning rate can be adjusted or the iteration times can be increased appropriately to ensure that the parameter adjustment can make the prediction result of the model closer to the measured value.

[0084] During the parameter fine-tuning process, attention should be paid to the coordination of the adjustment of the parameters of each layer to avoid excessive adjustment of the parameters of a layer affecting the feature extraction effect of other layers. At the same time, in order to prevent the model from overfitting the training set data during training, validation set data can be introduced regularly during training for testing, and the performance of the model on the validation set can be observed. If the loss function of the model on the training set continues to decrease, but the loss function on the validation set starts to rise, it means that the model has overfitting, and the complexity of the model needs to be reduced or regularized processing needs to be added. Through the above unsupervised pre-training and parameter fine-tuning process, the deep belief network model can fully learn the correlation between the features in the training set and the measured mechanical property data, and form a stable prediction ability.

[0085] Example 3: see Figure 4 The yield strength and tensile strength parameters in the smoothed force value displacement data are fused with the peak frequency and amplitude decay rate parameters in the spectral feature parameters to construct a first multi-dimensional feature matrix, and the specific process is as follows:

[0086] The yield strength parameter and the tensile strength parameter in the smooth force value displacement data are normalized to obtain standardized mechanical parameters. The normalization process is as follows: first, the maximum and minimum values of the yield strength and the tensile strength of all samples in the training set are counted, and then the yield strength and the tensile strength of each sample are converted according to a unified rule to make the converted parameter values within a specific interval. Through this processing, yield strength and tensile strength parameters of different orders of magnitude can be compared on the same numerical scale, avoiding the influence of the order of magnitude of the original parameters on subsequent feature fusion and model training.

[0087] The peak frequency parameter and the amplitude decay rate parameter in the spectral feature parameter are standardized to obtain standardized spectral parameters. The standardization processing method is the same as the normalization processing of yield strength and tensile strength, that is, the maximum and minimum values of the peak frequency and the amplitude decay rate in the training set are determined, and then the peak frequency and the amplitude decay rate of each sample are converted to the same numerical interval as the standardized mechanical parameters according to the same rule. In this way, the standardized mechanical parameters and the standardized spectral parameters remain consistent in the numerical range, providing a unified basis for subsequent feature splicing.

[0088] The standardized mechanical parameters and the standardized spectral parameters are spliced by column to form an initial feature matrix. When splicing, for each titanium alloy wire sample, its corresponding standardized yield strength, standardized tensile strength, standardized peak frequency, and standardized amplitude decay rate are arranged in order to form a row of data, and each data is taken as an element in the matrix. The row data of all samples are combined to form a two-dimensional matrix, which is the initial feature matrix. The number of rows of the initial feature matrix is equal to the number of samples, and the number of columns is equal to the total number of standardized parameters. Each element represents the value of a sample on a certain standardized parameter.

[0089] An independent component analysis algorithm is used to reduce the dimension of the initial feature matrix to remove redundant features and obtain an optimized multi-dimensional feature matrix, which is the first multi-dimensional feature matrix. The specific steps are as follows: the initial feature matrix is centered, the average value of each feature in all samples is calculated, and then the feature value of each sample is subtracted by the corresponding average value to make the average value of each feature after processing zero. Centering can eliminate the numerical bias of the features, making the comparison between different features more targeted.

[0090] The covariance matrix of the centered initial feature matrix is calculated, and each element in the covariance matrix reflects the linear correlation between two different features. The larger the value, the stronger the correlation. The eigenvalue decomposition of the covariance matrix is performed to obtain a plurality of eigenvalues and corresponding eigenvectors. The size of the eigenvalue represents the amount of information contained in the corresponding eigenvector. The larger the eigenvalue, the more information the corresponding eigenvector carries.

[0091] The eigenvalues are sorted according to their sizes from large to small, and then the ratio of the cumulative sum of the first k eigenvalues to the total sum of all eigenvalues, i.e. the cumulative contribution rate, is calculated. The first k eigenvectors with a cumulative contribution rate reaching a preset threshold are selected, and these eigenvectors are combined to form a projection matrix. The preset threshold is set by comprehensively considering the integrity of feature retention and the efficiency of dimension reduction, and by selecting a suitable k value, the main information in the initial features can be retained, and the dimension of the features can be reduced.

[0092] The initial feature matrix after the centering processing is multiplied by the projection matrix to obtain a multi-dimensional feature matrix after dimension reduction, i.e. an optimized multi-dimensional feature matrix, which is a first multi-dimensional feature matrix.

[0093]

[0094] wherein, represents the multi-dimensional feature matrix after dimension reduction, represents the initial feature matrix after the centering processing, represents the projection matrix.

[0095] The number of rows of the multi-dimensional feature matrix after dimension reduction is the same as that of the initial feature matrix, and the number of columns is k, i.e. the number of selected eigenvectors. Through this process, the redundant information existing in the initial feature matrix is removed, the feature dimension is reduced, and at the same time, the key information important for predicting the performance of titanium alloy wire is retained. When the optimized multi-dimensional feature matrix is input into the deep belief network model, the calculation amount of the model can be reduced, and the prediction speed of the model can be improved.

[0096] It can be understood that when the initial deep belief network model is trained by using the multi-dimensional feature matrix corresponding to the historical titanium alloy wire sample, the multi-dimensional feature matrix after dimension reduction corresponding to the historical titanium alloy wire sample can also be obtained through the foregoing steps, and the initial deep belief network model is trained by using the multi-dimensional feature matrix after dimension reduction corresponding to the historical titanium alloy wire sample, so as to reduce the calculation amount of the model, improve the training efficiency of the model, and at the same time, avoid the risk of overfitting of the model caused by redundant features.

[0097] In the entire feature fusion and dimension reduction process, it is necessary to ensure that all processing steps are based on the parameters in the training set to ensure the consistency of the processing rules. For the feature parameters of the newly collected titanium alloy wire sample (i.e. the titanium alloy wire to be detected), the same processing as determined by the maximum value, the minimum value and the projection matrix of the training set is also required, so that the first multi-dimensional feature matrix of the new sample and the feature matrix during the training of the model remain consistent in format and scale, and the model can accurately perform prediction processing.

[0098] Example 4:Figure 5 The echo signal data is subjected to a fast Fourier transform, and the amplitude spectrum and phase spectrum of the signal are extracted as spectral feature parameters, as follows:

[0099] The echo signal data is subjected to a truncation process to remove the noise segments at the beginning and end of the signal, obtaining an effective signal segment. During the acquisition process, due to electromagnetic interference when the detection device is started and stopped, the beginning and end of the signal often contain a lot of noise, which will affect the subsequent signal analysis accuracy. When processing, first observe the time-domain waveform of the echo signal, identify the beginning and end of the signal whose amplitude is significantly lower than the middle stable segment, and determine these parts as noise segments and remove them. For example, when the time-domain waveform of the echo signal presents a rapid fluctuation in amplitude at both ends and a low overall amplitude, while the middle part has a stable amplitude and a small fluctuation, the middle stable part is retained as an effective signal segment.

[0100] A Hanning window is added to the effective signal segment to reduce spectral leakage. The effective signal segment is finite in length in the time domain, and direct Fourier transform will cause spectral leakage due to the sudden truncation of the signal, resulting in deviation in the frequency domain analysis result. The Hanning window is a window function that can smoothly transition the signal to zero at both ends. After multiplying the effective signal segment by the corresponding window function value of the Hanning window, the truncation of the signal in the time domain becomes smooth. In specific operation, the length of the Hanning window is determined according to the length of the effective signal segment to ensure that the lengths are consistent, and then each data point of the effective signal segment is multiplied by the corresponding window function value of the Hanning window to obtain a windowed signal. After windowing, the amplitude of the signal at both ends gradually approaches zero, reducing the sidelobe interference in the frequency domain caused by the sudden truncation of the signal.

[0101] The windowed signal is subjected to a fast Fourier transform to convert the time-domain signal to a frequency-domain signal. The time-domain signal reflects the change of signal amplitude with time, while the frequency-domain signal reflects the distribution characteristics of different frequency components in the signal. The fast Fourier transform converts the windowed time-domain discrete signal to a frequency-domain signal through mathematical operations, obtaining the amplitude and phase information corresponding to each frequency point. During the conversion process, the number of Fourier transforms is determined according to the length of the windowed signal to ensure that the frequency-domain signal after transformation can fully reflect the frequency characteristics of the original signal. After the transformation is completed, a three-dimensional frequency-domain signal data containing frequency, amplitude and phase can be obtained, where the frequency represents different frequency components contained in the signal, the amplitude represents the intensity of each frequency component, and the phase represents the phase shift of each frequency component.

[0102] The frequency corresponding to the maximum peak value of the amplitude spectrum extracted from the frequency domain signal is taken as the peak frequency parameter, and the frequency range at which the amplitude decays to half is taken as the amplitude decay rate parameter. The amplitude spectrum is a curve of the amplitude of the frequency domain signal varying with frequency. The maximum peak value point in the amplitude spectrum can be found by observing the amplitude spectrum, and the frequency corresponding to the maximum peak value point is the peak frequency parameter. For example, in the amplitude spectrum, if the amplitude of a certain frequency point is significantly higher than that of other frequency points and is the maximum value in the entire spectrum, the value of the frequency point is the peak frequency parameter. The extraction of the amplitude decay rate parameter is to find the two frequency points corresponding to the amplitude decay to half of the peak value after determining the maximum peak value, i.e., a low frequency point lower than the peak frequency and a high frequency point higher than the peak frequency. The difference between the two frequency points is the amplitude decay rate parameter. For example, when the peak frequency is 5 MHz, the low frequency point corresponding to the amplitude decay to half of the peak value is 4.8 MHz, and the high frequency point is 5.3 MHz, the amplitude decay rate parameter is 0.5 MHz.

[0103] During the entire processing process, attention should be paid to the relationship between the sampling frequency of the echo signal data and the frequency resolution of the frequency domain signal after Fourier transform. The higher the sampling frequency, the higher the frequency resolution of the frequency domain signal, the more precise the distinguishable frequency components, but at the same time, it will also increase the workload of data processing; if the sampling frequency is too low, it may not be able to accurately capture the high frequency components in the signal, resulting in incomplete frequency feature extraction. In addition, when extracting the peak frequency and amplitude decay rate, abnormal peak interference in the frequency domain signal should be excluded. These abnormal peaks may be caused by sudden noise in the signal acquisition process and are not the reflection of the characteristics of the titanium alloy wire itself. By comparing the frequency domain signals of multiple titanium alloy wires of the same specification, common peak characteristics can be identified to distinguish between effective peaks and abnormal peaks, ensuring that the extracted frequency spectrum characteristic parameters can truly reflect the internal structural characteristics of the titanium alloy wire.

[0104] Example 5: Collect real-time force value displacement data of the titanium alloy wire to be detected during the tensile test process and echo signal data returned by the ultrasonic detection device, as follows:

[0105] The real-time force value data and displacement data of the titanium alloy wire during the tensile test are synchronously collected by sensors. The tensile test is carried out on a dedicated tensile testing machine, and the two ends of the titanium alloy wire are fixed on the clamps of the testing machine. With the movement of the clamps, the wire is subjected to axial tension and deformation. The real-time force value data is obtained by a tensile and compressive force sensor installed on the clamp. The sensor is consistent with the force direction of the wire and can sense the tension of the wire in real time and convert the mechanical signal into an electrical signal output. The displacement data is collected by a laser range finder, which is fixed on one side of the testing machine. The laser emission direction of the laser range finder is parallel to the tensile direction of the wire. The laser beam directly irradiates on the surface of the wire close to the clamp or the moving part connected with the clamp. The displacement change is calculated by measuring the time difference between the laser emission and reflection.

[0106] The ultrasonic detection device is controlled to emit ultrasonic pulses at a preset frequency to different radial sections of the titanium alloy wire, and receive the echo signal data returned by each section. The ultrasonic detection device is composed of an ultrasonic probe, a pulse generator and a signal receiver. The ultrasonic probe is installed on a mechanical arm that can move along the length direction of the wire, and can scan and detect multiple radial sections of the wire during the tensile test. The determination of the preset frequency needs to consider the diameter specification of the titanium alloy wire. For smaller diameter wires, a higher ultrasonic frequency is used. At this time, the wavelength of the ultrasonic pulse is shorter, which can identify smaller defects or structural changes in the wire. For larger diameter wires, the ultrasonic frequency is appropriately reduced to increase the propagation depth of the ultrasonic pulse in the wire, ensuring that the characteristics of the central region of the wire can be detected. During the detection process, the mechanical arm moves the ultrasonic probe according to the preset step. After moving to a preset position, the ultrasonic pulse is emitted to the radial section at the position. After the pulse passes through the wire, it is reflected by the opposite reflecting surface to form an echo signal. The echo signal is received by the ultrasonic probe and converted into an electrical signal, which is transmitted to the signal receiver for preliminary processing.

[0107] The time stamp label is generated by the system clock, which is accurate to the millisecond level. Each data point is automatically labeled with the corresponding time stamp when it is collected. For real-time force value data and displacement data, since they are collected synchronously by the sensor, each time point corresponds to a set of force value and displacement data. For echo signal data, since there is a time interval between the detection of different radial sections by the ultrasonic detection device, each echo signal data point has a time stamp of the time when it was collected. Through the time stamp label, the force value data, displacement data, and echo signal data at the same time can be associated, for example, when the wire is stretched to a certain displacement, the corresponding force value and the echo signal characteristics of the radial section at that time can form a complete set of associated data. The establishment of this time correlation index enables accurate analysis of the characteristic changes of the internal structure of the wire under specific mechanical conditions during subsequent data processing, providing a corresponding relationship in the time dimension for the fusion of force value displacement data and echo signal data.

[0108] During data collection, the synchronization between devices needs to be ensured to avoid deviations in the time stamp label due to device response delays. The hardware synchronization trigger mechanism can be used to start data collection for the tensile-compressive force sensor, laser range finder, and ultrasonic detection device under the same trigger signal, reducing the synchronization error between devices. At the same time, the collected data is cached in real time to prevent data loss due to slow data transmission speed. For the ultrasonic detection device, the transmission frequency and probe position need to be calibrated regularly to ensure consistent detection conditions for different radial sections and reduce the impact of device errors on echo signal data. In addition, before the tensile test begins, the sensor and ultrasonic detection device need to be preheated and calibrated to check the working status of the equipment and ensure that the collected data is accurate and reliable. Through the above data collection process, complete mechanical change data and internal structure characteristic data of the titanium alloy wire during the tensile test can be obtained, providing original data support for subsequent performance testing.

[0109] It should be noted that, in this text, relationship terms such as first and second are only used to distinguish one entity or operation from another, and do not necessarily require or imply any actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment.

[0110] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction and arrangement described. It is intended, therefore, to cover all modifications and adaptions that fall within the scope of the claims and their equivalents.

Claims

1. A method of detecting the properties of a titanium alloy wire, characterized by, The method comprises the steps of: Collecting real-time force value displacement data of titanium alloy wire to be detected during a tensile test and echo signal data returned by an ultrasonic detection device, and transmitting the real-time force value displacement data and the echo signal data to a data processing unit; Performing denoising processing on the real-time force value displacement data by the data processing unit using an adaptive median filtering algorithm to obtain smoothed force value displacement data, and performing fast Fourier transform on the echo signal data to extract amplitude spectrum and phase spectrum of the signal as frequency spectrum characteristic parameters; Fusing yield strength and tensile strength parameters in the smoothed force value displacement data with peak frequency and amplitude decay rate parameters in the frequency spectrum characteristic parameters to construct a first multi-dimensional feature matrix; Inputting the first multi-dimensional feature matrix into a pre-trained deep belief network model to perform prediction processing on the elongation and hardness value of the titanium alloy wire to be detected by the deep belief network model to obtain a mechanical property prediction result; Comparing the mechanical property prediction result with a preset performance threshold range, marking as unqualified when the mechanical property prediction result exceeds the preset performance threshold range, and generating a titanium alloy wire performance detection report containing unqualified items and corresponding parameters; The method comprises the steps of: Performing truncation processing on the echo signal data to remove noise segments at the beginning and end of the signal to obtain an effective signal segment; Adding a Hanning window to the effective signal segment to reduce frequency spectrum leakage to obtain a windowed signal; Performing fast Fourier transform on the windowed signal to convert time domain signal to frequency domain signal; Extracting the frequency corresponding to the maximum peak value of the amplitude spectrum as the peak frequency parameter, and calculating the frequency range when the amplitude decays to half as the amplitude decay rate parameter.

2. The method for detecting the performance of titanium alloy wire according to claim 1, characterized in that, The method comprises the steps of: Obtaining multi-dimensional feature matrices and measured mechanical property data of historical titanium alloy wire samples from a database, and dividing the historical titanium alloy wire samples into a training set and a validation set; Using the training set to perform unsupervised pre-training on an initial deep belief network model, extracting features layer by layer through stacking a restricted Boltzmann machine, and fine-tuning parameters of the initial deep belief network model using a back propagation algorithm; Using the validation set to perform performance evaluation on the fine-tuned deep belief network model, calculating a prediction error, and stopping training when the prediction error is less than a preset error threshold to obtain the pre-trained deep belief network model; Inputting the first multi-dimensional feature matrix into the pre-trained deep belief network model to output elongation prediction value and hardness value prediction value of the titanium alloy wire as the mechanical property prediction result.

3. The method of claim 2, wherein the titanium alloy wire is a wire having a diameter of 0.1 mm or less. The unsupervised pre-training of the initial deep belief network model by using the training set, the feature is extracted layer by layer by stacking the restricted Boltzmann machine, and the parameters of the initial deep belief network model are fine-tuned by using the back propagation algorithm, comprising: The number of neurons of the input layer, the hidden layer and the output layer of the initial deep belief network model is initialized, and the learning rate and the iteration number of the restricted Boltzmann machine are set; The multi-dimensional feature matrix in the training set is input into the first layer of the restricted Boltzmann machine, and the first layer feature mapping is obtained by training by using the contrastive divergence algorithm; The first layer feature mapping is taken as the input of the second layer of the restricted Boltzmann machine for training, and is sequentially stacked to the preset number of layers, and the unsupervised pre-training process is completed; The measured mechanical property data of the training set is taken as a label, the mean square error of the predicted value and the label is calculated by using the back propagation algorithm, the connection weight and the bias value between the neurons of each layer are adjusted according to the mean square error, and the parameter fine-tuning is completed.

4. The method of claim 3, wherein the titanium alloy wire is a wire having a diameter of 0.1 mm or less. The measured mechanical property data of the training set is taken as a label, the mean square error of the predicted value and the label is calculated by using the back propagation algorithm, the connection weight and the bias value between the neurons of each layer are adjusted according to the mean square error, and the parameter fine-tuning is completed. The multi-dimensional feature matrix of the training set is input into the deep belief network model which has completed the unsupervised pre-training, and the predicted mechanical property data of each sample is obtained; The mean square error between the predicted mechanical property data and the measured mechanical property data is calculated, and the mean square error is taken as a loss function; The gradient descent method is used to derive the loss function to obtain the gradient direction of the connection weight and the bias value of each layer; The connection weight and the bias value are updated according to the gradient direction with a preset learning rate, and the iteration is repeated until the loss function converges to a preset minimum value range.

5. The method of claim 1, wherein the titanium alloy wire is a wire having a diameter of 0.1 mm or less. The yield strength and tensile strength parameters in the smoothed force value displacement data are fused with the peak frequency and amplitude attenuation rate parameters in the spectral feature parameters to construct a first multi-dimensional feature matrix, comprising: The yield strength and tensile strength parameters in the smoothed force value displacement data are normalized to obtain standardized mechanical parameters; The peak frequency and amplitude attenuation rate parameters in the spectral feature parameters are standardized to obtain standardized spectral parameters; The standardized mechanical parameters and the standardized spectral parameters are spliced by column to form an initial feature matrix; An independent component analysis algorithm is used to reduce the dimension of the initial feature matrix to remove redundant features to obtain the first multi-dimensional feature matrix.

6. The method of claim 5, wherein the titanium alloy wire is a wire having a diameter of 0.1 mm or less. The independent component analysis algorithm is used to reduce the dimension of the initial feature matrix to remove redundant features to obtain the first multi-dimensional feature matrix, comprising: The initial feature matrix is centered to make the mean value of each feature zero; The covariance matrix of the centered initial feature matrix is calculated, and the eigenvalue and eigenvector of the covariance matrix are obtained by eigenvalue decomposition; According to the size of the eigenvalue, the first k eigenvectors with a cumulative contribution rate reaching a preset threshold are selected to form a projection matrix; The initial feature matrix after the centralization processing is multiplied by the projection matrix to obtain the first multi-dimensional feature matrix.

7. The method of claim 1, wherein the titanium alloy wire is a wire having a diameter of 0.1 mm or less. The real-time force value displacement data of the titanium alloy wire to be detected in the tensile test process and the echo signal data returned by the ultrasonic detection device are collected, and the echo signal data returned by the ultrasonic detection device includes: The real-time force value data and displacement data of the titanium alloy wire in the tensile test are synchronously collected by a sensor, the real-time force value data is obtained by a tension and compression force sensor, and the displacement data is collected by a laser range finder; The ultrasonic detection device is controlled to emit ultrasonic pulses at a preset frequency to different radial sections of the titanium alloy wire, and receive echo signal data returned by each section, and the preset frequency is dynamically adjusted according to the diameter specification of the titanium alloy wire. Timestamp labels are added to the real-time force value data, displacement data and echo signal data to establish a time correlation index of multi-source data.

8. A titanium alloy wire performance detection system for implementing the titanium alloy wire performance detection method according to any one of claims 1 to 7, characterized in that, It includes: The acquisition module is used for collecting the real-time force value displacement data of the titanium alloy wire to be detected in the tensile test process and the echo signal data returned by the ultrasonic detection device, and transmitting the real-time force value displacement data and the echo signal data to the data processing unit; The preprocessing module is used for denoising the real-time force value displacement data by the data processing unit to obtain smooth force value displacement data by using an adaptive median filtering algorithm, and simultaneously executing fast Fourier transform on the echo signal data to extract spectral feature parameters; The fusion module is used for performing feature layer fusion on the yield strength and tensile strength parameters in the smooth force value displacement data and the peak frequency and amplitude decay rate parameters in the spectral feature parameters to construct a first multi-dimensional feature matrix; The prediction module is used for inputting the first multi-dimensional feature matrix into a pre-trained deep belief network model, and performing prediction processing on the elongation and hardness value of the titanium alloy wire by the deep belief network model to obtain a mechanical property prediction result; The report generation module is used for comparing the mechanical property prediction result with a preset performance threshold range, marking as unqualified when the mechanical property prediction result exceeds the preset performance threshold range, and generating a titanium alloy wire performance detection report containing unqualified items and corresponding parameters.

9. A computing device, comprising: The storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to realize the titanium alloy wire performance detection method in any one of claims 1 to 7.

Images