CN115494302A - Resolution bandwidth intelligent setting method and system for spectrum testing of aerospace products - Google Patents

Abstract

A resolution bandwidth intelligent setting method and system for aerospace product spectrum test comprises the following steps: determining main factors of automatic configuration of a frequency spectrum project according to common parameters needing to be configured in a product frequency spectrum test project test; extracting influence factors set by resolution bandwidth according to the historical test data characteristics of the product spectrum test project, and constructing a feature vector; establishing a neural network model with resolution bandwidth setting, and training the neural network model by using a training sample set to obtain a resolution bandwidth prediction model; and setting the resolution bandwidth of the spectrum test item by using the resolution bandwidth prediction model. According to the method, the influence factors of the frequency spectrum project test resolution bandwidth setting are extracted according to the historical test data characteristics of the frequency spectrum project of the product, the advantages of a machine learning method in parameter optimization are utilized, a neural network technology is used for fitting a nonlinear model of the resolution bandwidth and the influence factors thereof, and the intelligent setting of the frequency spectrum project test resolution bandwidth is realized.

Description

Resolution bandwidth intelligent setting method and system for spectrum testing of aerospace products

technical field

The invention relates to an intelligent setting method of resolution bandwidth for spectrum testing of aerospace products, and belongs to the technical field of automatic testing of electrical properties.

Background technique

With the increasing maturity of digital and intelligent technologies and the upgrading of aerospace product test systems, most of the spectrum test items can be automated in the ground test verification stage of aerospace products, and the test efficiency has been greatly improved.

When the existing test system tests spectrum items, the test engineer needs to manually test once according to the actual situation of the stand-alone machine and the performance of the instrument, select the appropriate parameters, and then input the parameters into the test system. However, there are many types and few quantities of aerospace products, and different products need to manually set the test parameters of spectrum items, which is not conducive to the further improvement of test efficiency. In addition, there are many test procedures and projects for aerospace products, and the resources of instruments and equipment are limited. After the replacement of instruments and equipment in different test procedures, parameter reuse may cause parameter mismatch, resulting in inaccurate test data and unable to truly reflect the stand-alone performance indicators. It leads to a second test, resulting in a waste of time and resources, which is not conducive to schedule control and cost control.

Contents of the invention

The technical solution problem of the present invention is: to overcome the deficiencies of the prior art, to provide a resolution bandwidth intelligent setting method for the spectrum test of aerospace products, and to use machine learning technology to establish a neural network that conforms to the resolution bandwidth setting of the spectrum test of aerospace products network model. Train the model according to the accumulated data of the test, fit the nonlinear relationship between the resolution bandwidth and the characteristics of the measured signal and other parameters of the instrument, and find the optimal setting of the resolution bandwidth, so as to realize the automatic testing of spectrum items and further improve Product testing efficiency.

Technical solution of the present invention is:

An intelligent resolution bandwidth setting method for spectrum testing of aerospace products, comprising:

Determine the main factors for the automatic configuration of spectrum items according to the common parameters that need to be configured for product spectrum test items;

According to the characteristics of the historical test data of the product spectrum test project, extract the influencing factors of the resolution bandwidth setting, and construct the feature vector;

Establish a neural network model with resolution and bandwidth settings, use the training sample set to train the neural network model, and obtain a resolution and bandwidth prediction model;

The resolution bandwidth of the spectrum test item is set by using the resolution bandwidth prediction model.

Further, the common parameters that need to be configured for the product spectrum test item test include: carrier frequency, bandwidth, built-in attenuation, reference level, amplitude resolution, resolution bandwidth, and video bandwidth;

Carrier frequency and bandwidth are the input conditions of the test;

The settings of built-in attenuation, reference level, and amplitude resolution are directly affected by the output power of the measured signal, and the parameter values and the output power of the measured signal are linear functions;

The built-in attenuation is used to protect the mixer inside the spectrum analyzer, and the minimum value of the built-in attenuation is equal to the maximum output power of the signal under test minus the maximum input power value of the mixer;

The reference level is equal to the maximum output power of the signal under test plus b, b>0;

The amplitude resolution is equal to the maximum output power of the signal under test minus the minimum output power of the signal under test, and then divided by 10;

Video bandwidth and resolution bandwidth are set in a fixed multiple relationship.

Further, the main factor of the automatic configuration of the spectrum test items is the resolution bandwidth.

Further, the extracted influencing factors of resolution bandwidth setting include: center frequency f _c of the signal under test, bandwidth s, reference level l, attenuation a, sweep time t and index requirements.

Further, taking the influencing factors of the resolution bandwidth as the input vector factor of the neural network model for setting the resolution bandwidth, the input feature vector is expressed as:

X = [f _c slati]

Normalize the eigenvectors to obtain normalized eigenvectors; collect historical test data and formulate data eigenvectors X _k , k=1,2,...,m, where m is the maximum number of samples; and mark the samples , set the label value, then the vector X _k and its corresponding label are used as the sample set of the neural network model, 70% of the sample set is taken as the training sample set, and the remaining 30% is used as the test sample set.

Further, the normalized feature vector, the calculation formula is:

Among them, f _min and f _max are the test frequency of the spectrum analyzer, s _min and s _max are the test bandwidth of the spectrum analyzer, l _min and l _max are the reference levels of the spectrum analyzer, a _min and a _max are the internal attenuation of the spectrum analyzer, t _min and t _max is the scanning time of the spectrum analyzer, and i _min and i _max are the common index requirements of the traveling wave tube amplifier.

Further, establish a neural network model with resolution bandwidth setting, use the training sample set to train the neural network model, and obtain a resolution bandwidth prediction model, specifically:

Construct a fully connected neural network, that is, a neural network model with resolution and bandwidth settings. The network model structure includes an input layer, 3 hidden layers, and an output layer; according to the constructed feature vector, the dimension of the input layer vector is 6, and the output The layer uses softmax classification, and the output is the setting probability of 10 types of resolution bandwidth; input the training sample set into the neural network model, set the number of training iterations, the number of batch samples and the learning rate, and continuously train and optimize the network model to obtain the resolution bandwidth predictive model.

Further, the use of the resolution bandwidth prediction model to set the resolution bandwidth of the spectrum test item is specifically: embedding the obtained resolution bandwidth prediction model into the product test system, and reading the parameters of the measured signal center frequency f _c , bandwidth s, automatically identify the parameters reference level l, attenuation a, scan time t and index requirement i, generate real-time feature vectors, obtain the setting probabilities of 10 preset resolution bandwidths, and take the maximum value of the probabilities.

Further, the present invention also proposes an intelligent resolution bandwidth setting system for spectrum testing of aerospace products, including:

The main factor determination module: according to the common parameters that need to be configured for the product spectrum test item test, determine the main factors for the automatic configuration of the spectrum item; the common parameters that need to be configured for the product spectrum test item test include: carrier frequency, bandwidth, built-in attenuation, reference voltage level, amplitude resolution, resolution bandwidth, video bandwidth; carrier frequency and bandwidth are the input conditions of the test; the settings of built-in attenuation, reference level, and amplitude resolution are directly affected by the output power of the signal under test. The output power of the measured signal is a linear function; the built-in attenuation is used to protect the mixer inside the spectrum analyzer, and the minimum value of the built-in attenuation is equal to the maximum output power of the signal under test minus the maximum input power value of the mixer; the reference level is equal to the The maximum output power of the signal plus b, b>0; the amplitude resolution is equal to the maximum output power of the signal under test minus the minimum output power of the signal under test, and then divided by 10; the video bandwidth and resolution bandwidth set a fixed multiple relationship; spectrum test The main factor of project automatic configuration is resolution bandwidth;

Eigenvector construction module: according to the characteristics of historical test data of product spectrum test items, extract the influencing factors of resolution bandwidth setting, and construct the eigenvector; the extracted influencing factors of resolution bandwidth setting include: the center frequency f _c of the signal under test, the bandwidth s, Reference level l, attenuation a, scan time t and index requirements;

Taking the influencing factors of resolution bandwidth as the input vector factor of setting the resolution bandwidth neural network model, the input feature vector is expressed as:

X = [f _c slati]

Neural network model establishment and training module: establish a neural network model with resolution and bandwidth settings, use the training sample set to train the neural network model, and obtain a resolution and bandwidth prediction model;

Resolution bandwidth setting module: use the resolution bandwidth prediction model to set the resolution bandwidth of the spectrum test item.

The beneficial effect of the present invention compared with prior art is:

(1) The present invention uses machine learning technology to establish a neural network model that meets the resolution bandwidth setting of aerospace product spectrum testing. Train the model according to the accumulated data of the test, fit the nonlinear relationship between the resolution bandwidth and the characteristics of the measured signal and other parameters of the instrument, and find the optimal setting of the resolution bandwidth, so as to realize the automatic testing of spectrum items and further improve Product testing efficiency.

(2) According to the characteristics of historical test data and engineering practice of aerospace product spectrum items, the present invention innovatively utilizes the advantages of machine learning methods in parameter optimization, and uses neural network technology to fit the nonlinear model of resolution bandwidth and its influencing factors, Realized the intelligent setting of spectrum project test resolution bandwidth.

(3) The present invention embeds the trained neural network model into the aerospace product testing system testing system, intelligently sets the resolution bandwidth of the aerospace product spectrum item, eliminates the cumbersome manual setting of spectrum test parameters, and avoids parameter failures when the spectrum analyzer is not used. Reuse the introduced test errors to ensure consistent test baselines and further improve test efficiency.

Description of drawings

Fig. 1 is a flow chart of the method of the present invention;

Figure 2 is a structural diagram of the neural network model.

detailed description

Specific embodiments of the present invention will be further described in detail below in conjunction with the accompanying drawings.

The present invention has carried out research on parameter setting of spectrum test items. Through a series of experiments, the commonly used parameters that need to be configured for spectrum testing, including built-in attenuation, reference level and amplitude resolution, resolution bandwidth, video bandwidth, etc., are analyzed and found Spectrum projects automatically configure the main factor resolution bandwidth.

The present invention will take the test of an aerospace product traveling wave tube amplifier as an example, and illustrate the specific process of the present invention in conjunction with FIG. 1 .

Step 1: Determine the main factors for the automatic configuration of the spectrum project

Common parameters that need to be configured for wave tube amplifier spectrum project testing are shown in Table 1:

Table 1 Commonly used configuration parameters for spectrum project testing

Including carrier frequency, bandwidth, built-in attenuation, reference level and amplitude resolution, resolution bandwidth, video bandwidth, etc. Among them, carrier frequency and bandwidth are the input conditions of the test; parameters such as built-in attenuation, reference level and amplitude resolution are directly affected by the output power of the measured signal, and there is a linear functional relationship between them, which can be automatically identified and calculated by the test system; The video bandwidth can be set to a fixed multiple relationship with the resolution bandwidth; while the resolution bandwidth needs to comprehensively consider the characteristics of the measured signal, test parameters, instrument noise floor, scan time and the performance of the instrument itself, and there is a nonlinear functional relationship with the above factors. It is the main factor for automatic configuration of spectrum test items.

Step 2: Extract the influencing factors of the resolution bandwidth setting and construct the feature vector

According to the historical test data characteristics of the traveling wave tube amplifier spectrum project, Keysight's 9030B-50G is selected as the test instrument, and the factors affecting the resolution bandwidth setting include: the center frequency f _c of the tested signal, the bandwidth s, the reference level l, and the attenuation a , scan time t and index requirement i. Table 2 below is an example diagram of the data statistics of the influencing factors.

Table 2 Statistical example of factors affecting resolution bandwidth setting

The influencing factors of the resolution bandwidth are used as the input vector factors for setting the resolution bandwidth neural network model, then the input feature vector is expressed as:

X＝[f _c slati] (1)

In order to weaken the influence of different feature units and scales, the features are normalized to obtain a normalized feature vector. The calculation formula is:

Among them, f _min and f _max are the test frequencies of the spectrum analyzer, taking the minimum value of 3Hz and the maximum value of 50GHz respectively; s _min and s _max are the test bandwidth of the spectrum analyzer, taking the minimum value of 1Hz and the maximum value of 25GHz respectively; l _min and l _max are Spectrum analyzer reference level, take the minimum value -25dBm and maximum value 25dBm respectively, a _min and a _max are the internal attenuation of the spectrum analyzer, respectively take the minimum value -80dBm and maximum value 0dBm, t _min and t _max are the scan time of the spectrum analyzer, Take the minimum value of 0.1ms and the maximum value of 60s, respectively, i _min and i _max are common index requirements for traveling wave tube amplifiers, and take the minimum value of -80dBc and maximum value of -35dBc respectively.

Collect the historical test data of the TWT amplifier, and formulate the data feature vector X _k (k=1,2,...,m), where m is the maximum number of samples. And mark the sample, set the label value, then the vector X _k and its corresponding label are used as the sample set of the neural network model, 70% of the sample set is taken as the training sample set, and the remaining 30% is used as the test sample set.

Step 3: Use the training sample set to train the neural network model to obtain the resolution bandwidth prediction model

Construct a fully connected neural network. The network model structure is shown in Figure 2, including an input layer, 3 hidden layers, and an output layer. The network training parameter settings are shown in Table 3.

According to the feature vector constructed in step 2, the input layer vector dimension is 6, the output layer uses softmax classification, and the output is 10 types of resolution bandwidth (3Hz, 10Hz, 30Hz, 100Hz, 300Hz, 1kHz, 3kHz, 10kHz, 30kHz, 100kHz ) setting probability. The calculation formula is:

为softmax输出的概率。Among them, X is the input vector, W _10×6 is the weight, b is the bias item,

Probability for softmax output.

Then the probability of each type of resolution bandwidth is:

和真实数据Y一定会存在差异，所以建立分辨率带宽设置的神经网络模型的关键是找到合适的权重值W，使模型能准确反应分辨率带宽设置值与分辨率带宽影响因素之间的非线性关系。将影响因素的信息输入到神经网络模型中，设置训练迭代次数，批处理样本数量和学习率，得到神经网络的输出

在损失函数上采取交叉熵损失去函数评估Y与

之间分布的差异，不断训练优化网络模型，对权重值W进行调节，使

无限逼近Y。当它们之间的误差达到10^-3时，停止训练。The output of the neural network

There must be differences with the real data Y, so the key to establishing a neural network model with resolution bandwidth setting is to find an appropriate weight value W, so that the model can accurately reflect the nonlinearity between the resolution bandwidth setting value and the resolution bandwidth influencing factors relation. Input the information of influencing factors into the neural network model, set the number of training iterations, the number of batch samples and the learning rate, and get the output of the neural network

Take the cross-entropy loss on the loss function to evaluate the function Y and

The distribution difference between them, continuously train and optimize the network model, adjust the weight value W, so that

infinitely close to Y. When the error between them reaches 10 ^-3 , the training is stopped.

Table 3 Network training parameter setting table

Step 4: Use the prediction model to set the resolution bandwidth of the spectrum test item

Embed the prediction model obtained in step 3 into the TWT amplifier test system, read the parameter table of the tested product, obtain the center frequency f, bandwidth s, automatically identify the parameter reference level l, attenuation a, scan time t and index requirement i, Generate the real-time feature vector x, obtain the set probability y=softmax(W ^T x+b) of the preset 10 types of resolution bandwidth, and take the resolution bandwidth with the highest probability as the input of the test system.

In this embodiment, the test system embedded in the model is run, and a total of more than 100 traveling wave tube amplifiers covering the L, S, X, and Ka frequency bands are measured, compared with the manual preset spectrum parameter method, and 10 different spectrums are tested with each single machine Figure comparison.

Table 4 Comparison of different spectrums

It can be seen from Table 4 that the test system embedded with the neural network model saves about 90% of the preset time, avoids the test error caused by parameter multiplexing without using the spectrum analyzer, achieves the optimization purpose of the method, and realizes the frequency spectrum at the same time. Intelligent testing of the project.

The present invention uses machine learning to establish a suitable prediction model of the neural network model of resolution bandwidth, quickly and accurately predicts the resolution bandwidth of aerospace product spectrum testing, solves the problem of manual setting of resolution bandwidth, and realizes fully automatic testing of spectrum items. Technological progress in system intelligence.

The above is one of the best specific implementation modes of the present invention, and there may also be other various embodiments. The scope of protection of the present invention is not limited thereto. Any person familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present invention, and all should belong to the scope of protection of the appended claims of the present invention.

Parts not described in detail in the present invention belong to the common knowledge of those skilled in the art.

Claims (10)

1. An intelligent setting method for resolution bandwidth of aerospace product spectrum test is characterized by comprising the following steps:

determining main factors of automatic configuration of a frequency spectrum project according to common parameters needing to be configured in a product frequency spectrum test project test;

extracting influence factors set by resolution bandwidth according to the historical test data characteristics of the product spectrum test project, and constructing a feature vector;

establishing a neural network model with resolution bandwidth setting, and training the neural network model by using a training sample set to obtain a resolution bandwidth prediction model;

and setting the resolution bandwidth of the spectrum test item by using the resolution bandwidth prediction model.

2. The intelligent setting method for resolution bandwidth of aerospace product spectrum test as claimed in claim 1, wherein: the common parameters required to be configured for the product spectrum test item test comprise: carrier frequency, bandwidth, built-in attenuation, reference level, amplitude resolution, resolution bandwidth, video bandwidth;

carrier frequency and bandwidth are input conditions of the test;

the settings of built-in attenuation, reference level and amplitude resolution are directly influenced by the output power of the measured signal, and the parameter value and the output power of the measured signal are in a linear functional relationship;

the built-in attenuation is used for protecting a mixer in the frequency spectrograph, and the minimum value of the built-in attenuation is equal to the maximum output power of a measured signal minus the maximum input power value of the mixer;

the reference level is equal to the maximum output power of the measured signal plus b, b >0;

the amplitude resolution is equal to the maximum output power of the measured signal minus the minimum output power of the measured signal, and then divided by 10;

the video bandwidth and the resolution bandwidth are set to be in a fixed multiple relation.

3. The intelligent setting method for resolution bandwidth of aerospace product spectrum test as claimed in claim 2, wherein: the main factor for automatic configuration of the spectrum test item is the resolution bandwidth.

4. The intelligent setting method for resolution bandwidth of aerospace product spectrum test as claimed in claim 2, wherein: factors that influence the extracted resolution bandwidth setting include: measured signal center frequency f _c Bandwidth s, reference level l, attenuation a, scan time t, and index requirements.

5. The intelligent setting method for resolution bandwidth of aerospace product spectrum test of claim 4, wherein: taking the influence factor of the resolution bandwidth as an input vector factor of the neural network model for setting the resolution bandwidth, and expressing the input feature vector as follows:

X＝[f _c s l a t i]

normalizing the feature vector to obtain normalized feature vectorAn amount; collecting historical test data and making data characteristic vector X _k K =1,2, \ 8230;, m, where m is the maximum number of samples; and marking the sample, setting a label value, and then vector X _k And the labels corresponding to the training samples are used as a sample set of the neural network model, 70% of the sample set is extracted as a training sample set, and the rest 30% of the sample set is used as a testing sample set.

6. The method for intelligently setting the resolution bandwidth of the aerospace product spectrum test as claimed in claim 5, wherein: normalizing the feature vector, wherein the calculation formula is as follows:

wherein f is _min And f _max Testing frequency, s, for a spectrometer _min And s _max Testing the Bandwidth for a spectrometer,/ _min And l _max For spectrometer reference levels, a _min And a _max Internal attenuation of spectrometer, t _min And t _max For spectrometer scan time, i _min And i _max The method is a common index requirement of the traveling wave tube amplifier.

7. The method for intelligently setting the resolution bandwidth of the aerospace product spectrum test as claimed in claim 1, wherein: establishing a neural network model with resolution bandwidth setting, training the neural network model by using a training sample set, and obtaining a resolution bandwidth prediction model, wherein the method specifically comprises the following steps:

constructing a fully-connected neural network, namely a neural network model with resolution and bandwidth, wherein the network model structure comprises an input layer, 3 hidden layers and an output layer; according to the constructed feature vector, the vector dimension of an input layer is 6, the output layer is classified by softmax, and the set probability of 10 types of resolution bandwidth is output; and inputting the training sample set into a neural network model, setting training iteration times, batch processing sample number and learning rate, continuously training an optimization network model, and obtaining a resolution bandwidth prediction model.

8. The intelligent setting method for resolution bandwidth of aerospace product spectrum test as claimed in claim 7, wherein: the setting of the resolution bandwidth of the spectrum test item by using the resolution bandwidth prediction model specifically comprises the following steps: embedding the obtained resolution bandwidth prediction model into a product test system, and reading the central frequency f of the measured signal of the parameter _c And the bandwidth s is used for automatically identifying the parameter reference level l, the attenuation a, the scanning time t and the index requirement i, generating a real-time characteristic vector, obtaining the setting probability of the preset 10-class resolution bandwidth, and taking the maximum value of the probability.

9. The resolution bandwidth intelligent setting system for the aerospace product spectrum test is characterized by comprising the following components:

the main factor determining module: determining main factors of automatic configuration of a frequency spectrum project according to common parameters needing to be configured in a product frequency spectrum test project test; the common parameters required to be configured for the product spectrum test project test comprise: carrier frequency, bandwidth, built-in attenuation, reference level, amplitude resolution, resolution bandwidth, video bandwidth; carrier frequency and bandwidth are input conditions of the test; the setting of built-in attenuation, reference level and amplitude resolution is directly influenced by the output power of a measured signal, and the parameter value and the output power of the measured signal are in a linear functional relationship; the built-in attenuation is used for protecting a mixer in the frequency spectrograph, and the minimum value of the built-in attenuation is equal to the maximum output power of a measured signal minus the maximum input power value of the mixer; the reference level is equal to the maximum output power of the measured signal plus b, b >0; the amplitude resolution is equal to the maximum output power of the signal to be measured minus the minimum output power of the signal to be measured, and then divided by 10; setting a fixed multiple relation between the video bandwidth and the resolution bandwidth; the main factor of the automatic configuration of the spectrum test item is the resolution bandwidth;

the feature vector construction module: extracting influence factors set by resolution bandwidth according to the historical test data characteristics of the product spectrum test project, and constructing a feature vector; effect of extracted resolution Bandwidth settingsThe factors include: center frequency f of measured signal _c Bandwidth s, reference level l, attenuation a, scanning time t and index requirements;

taking the influence factor of the resolution bandwidth as an input vector factor of the neural network model for setting the resolution bandwidth, and expressing the input feature vector as follows:

X＝[f _c s l a t i]

the neural network model building and training module comprises: establishing a neural network model with resolution bandwidth setting, and training the neural network model by using a training sample set to obtain a resolution bandwidth prediction model;

a resolution bandwidth setting module: and setting the resolution bandwidth of the spectrum test item by using the resolution bandwidth prediction model.

10. The intelligent setting system for resolution bandwidth of aerospace product spectrum test of claim 9, wherein:

normalizing the feature vector to obtain a normalized feature vector; collecting historical test data and making data characteristic vector X _k K =1,2, \8230;, m, where m is the maximum number of samples; and marking the sample, setting a label value, and then vector X _k And the corresponding labels are used as a sample set of the neural network model, 70% of the sample set is extracted as a training sample set, and the rest 30% is used as a testing sample set;

normalizing the feature vector, wherein the calculation formula is as follows:

wherein f is _min And f _max Testing the frequency, s, for a spectrometer _min And s _max Testing the bandwidth for a spectrometer,/ _min And l _max For spectrometer reference levels, a _min And a _max For internal attenuation of the spectrometer, t _min And t _max For spectrometer scan time, i _min And i _max For the common index requirement of traveling wave tube amplifier；

Establishing a neural network model with resolution bandwidth setting, training the neural network model by using a training sample set, and obtaining a resolution bandwidth prediction model, wherein the method specifically comprises the following steps:

constructing a fully-connected neural network, namely a neural network model with resolution and bandwidth, wherein the network model structure comprises an input layer, 3 hidden layers and an output layer; according to the constructed feature vector, the vector dimension of an input layer is 6, the output layer is classified by softmax, and the set probability of 10 types of resolution bandwidth is output; inputting a training sample set into a neural network model, setting training iteration times, batch processing sample number and learning rate, continuously training an optimization network model, and obtaining a resolution bandwidth prediction model;

the setting of the resolution bandwidth of the spectrum test item by using the resolution bandwidth prediction model specifically comprises the following steps: embedding the obtained resolution bandwidth prediction model into a product testing system, and reading the central frequency f of the measured signal of the parameter _c And the bandwidth s is used for automatically identifying the parameter reference level l, the attenuation a, the scanning time t and the index requirement i, generating a real-time characteristic vector, obtaining the setting probability of the preset 10-class resolution bandwidth, and taking the maximum value of the probability.

Images