A Fault Detection Method for an Optoelectronic Accelerometer Driving Chip

Abstract

The present invention discloses a method for detecting faults in a driving chip of an optoelectronic accelerometer, belonging to the technical field of measuring electrical variables. First, the output signals of each module of the driving chip are divided according to the zero-crossing point to obtain positive and negative half-wave sets, and then the waveform symmetry value and the waveform symmetry stability value are extracted. The positive and negative half-wave sets are transformed into the frequency domain and divided into important frequencies and clutter frequencies to generate four frequency sets. By calculating the differences between adjacent modules on the four frequency sets, the positive half-wave frequency variation value and the negative half-wave frequency variation value of adjacent modules are obtained. At the same time, based on the differences between adjacent modules in the waveform symmetry value and the waveform symmetry stability value, the waveform symmetry difference and the waveform stability difference of adjacent modules are obtained. Finally, these characteristic values of each adjacent module are used as samples to be input into a convolutional neural network, so as to accurately obtain the fault value of the driving chip, significantly improving the fault detection accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of measuring electrical variables, and particularly relates to a method for detecting faults in a driving chip of an optoelectronic accelerometer. Background Art

[0002] Each module in the driving chip of the optoelectronic accelerometer successively includes a signal generation module, an operational amplifier module, and a coupling circuit module. The signal generation module generates a sine-wave modulation signal, which is amplified by the operational amplifier module, and the amplified signal is coupled by the coupling circuit module and then applied to the optoelectronic accelerometer.

[0003] Traditional methods for detecting faults in driving chips mainly rely on measuring simple electrical parameters, such as the voltage, current, resistance, etc. of the circuit. However, this detection method has obvious limitations for the complex fault modes of each module inside the chip, especially when the fault manifests as subtle changes in the signal waveform or abnormal frequency characteristics. For example, if the sine-wave modulation signal generated by the signal generation module has an asymmetric waveform or an increase in harmonic components, it is difficult to detect only by measuring electrical parameters. Therefore, the existing methods for detecting faults in driving chips have the problem of low fault detection accuracy. Summary of the Invention

[0004] Aiming at the above deficiencies in the prior art, a method for detecting faults in a driving chip of an optoelectronic accelerometer provided by the present invention solves the problem of low fault detection accuracy in existing driving chips.

[0005] In order to achieve the above object of the invention, the technical solution adopted by the present invention is: A method for detecting faults in a driving chip of an optoelectronic accelerometer, comprising:

[0006] Dividing the output signals of each module in the driving chip according to the zero-crossing point to obtain a positive half-wave set and a negative half-wave set;

[0007] For the positive half-wave set and the negative half-wave set of each module, extracting the waveform symmetry value and the waveform symmetry stability value;

[0008] Converting the positive half-wave set and the negative half-wave set to the frequency domain, dividing the frequency domain information into important frequencies and clutter frequencies to obtain four frequency sets, where the four frequency sets include: a positive half-wave important frequency set, a positive half-wave clutter frequency set, a negative half-wave important frequency set, and a negative half-wave clutter frequency set;

[0009] According to the differences between adjacent modules in the driving chip on the four frequency sets, obtaining the positive half-wave frequency variation value and the negative half-wave frequency variation value of adjacent modules;

[0010] Based on the differences in waveform symmetry values and waveform symmetry stability values between adjacent modules in the driver chip, the waveform symmetry difference and waveform stability difference of adjacent modules are obtained;

[0011] Taking the positive half-wave frequency variation value, negative half-wave frequency variation value, waveform symmetry difference, and waveform stability difference of each adjacent module as samples and inputting them into a convolutional neural network to obtain the driver chip fault value.

[0012] Furthermore, the process of extracting the waveform symmetry value is as follows:

[0013] For each positive half-wave in the positive half-wave set, calculate the area of the positive half-wave;

[0014] For each negative half-wave in the negative half-wave set, calculate the area of the negative half-wave;

[0015] Using the positive half-wave area and the peak value of the positive half-wave as elements, construct a positive half-wave feature vector;

[0016] Using the negative half-wave area and the absolute value of the peak value of the negative half-wave as elements, construct a negative half-wave feature vector;

[0017] Calculate the similarity between the positive half-wave feature vector and the negative half-wave feature vector in the same period;

[0018] Taking the average value of the similarities of each period to obtain the waveform symmetry value.

[0019] Furthermore, the formula for calculating the waveform symmetry stability value is: , where γ is the waveform symmetry stability value and σ is the standard deviation of the similarities of each period.

[0020] Furthermore, perform time-frequency conversion on each positive half-wave in the positive half-wave set and each negative half-wave in the negative half-wave set respectively to obtain positive half-wave frequency domain information and negative half-wave frequency domain information, and divide each frequency domain information into an important frequency set and a clutter frequency set to obtain a positive half-wave important frequency set, a positive half-wave clutter frequency set, a negative half-wave important frequency set, and a negative half-wave clutter frequency set.

[0021] Furthermore, the process of obtaining the positive half-wave important frequency set and the positive half-wave clutter frequency set is as follows:

[0022] Perform time-frequency conversion on each positive half-wave in the positive half-wave set respectively to obtain positive half-wave frequency domain information;

[0023] Taking the amplitudes greater than the amplitude threshold in the positive half-wave frequency domain information as positive half-wave important amplitudes, and taking the other amplitudes as positive half-wave clutter amplitudes;

[0024] Arrange the frequencies corresponding to the positive half-wave important amplitudes of all periods in sequence to construct a positive half-wave important frequency set;

[0025] Arrange the frequencies corresponding to the positive half-wave clutter amplitudes of all cycles in sequence to construct a positive half-wave clutter frequency set;

[0026] Alternatively, the process of obtaining the negative half-wave important frequency set and the negative half-wave clutter frequency set is as follows:

[0027] Perform time-frequency conversion on each negative half-wave in the negative half-wave set respectively to obtain negative half-wave frequency domain information, and take the absolute value of the amplitudes in the negative half-wave frequency domain information;

[0028] Take the amplitudes greater than the amplitude threshold in the negative half-wave frequency domain information as the negative half-wave important amplitudes, and take the other amplitudes as the negative half-wave clutter amplitudes;

[0029] Arrange the frequencies corresponding to the negative half-wave important amplitudes of all cycles in sequence to construct a negative half-wave important frequency set;

[0030] Arrange the frequencies corresponding to the negative half-wave clutter amplitudes of all cycles in sequence to construct a negative half-wave clutter frequency set.

[0031] Further, the process of obtaining the positive half-wave frequency variation value between adjacent modules and the negative half-wave frequency variation value between adjacent modules includes:

[0032] Obtain the first variation coefficient of the positive half-wave according to the gap between adjacent modules in the positive half-wave important frequency set in the driving chip; obtain the first variation coefficient of the negative half-wave according to the gap between adjacent modules in the negative half-wave important frequency set in the driving chip;

[0033] Obtain the second variation coefficient of the positive half-wave according to the gap between adjacent modules in the positive half-wave clutter frequency set in the driving chip; obtain the second variation coefficient of the negative half-wave according to the gap between adjacent modules in the negative half-wave clutter frequency set in the driving chip;

[0034] Perform weighting on the first variation coefficient of the positive half-wave and the second variation coefficient of the positive half-wave to obtain the positive half-wave frequency variation value between adjacent modules;

[0035] Perform weighting on the first variation coefficient of the negative half-wave and the second variation coefficient of the negative half-wave to obtain the negative half-wave frequency variation value between adjacent modules.

[0036] Further, the process of obtaining the first variation coefficient all includes:

[0037] Take the intersection of the important frequency sets of adjacent modules to obtain an important frequency intersection;

[0038] Exclude the important frequency intersection from the important frequency set of the subsequent module among adjacent modules to obtain new important frequencies;

[0039] Calculate the first variation coefficient according to the new important frequencies;

[0040] Alternatively, the process of obtaining the second variation coefficient all includes:

[0041] Take the intersection of the clutter frequency sets of adjacent modules to obtain the clutter frequency intersection;

[0042] Exclude the clutter frequency intersection from the clutter frequency set of the subsequent module among adjacent modules to obtain the newly added clutter frequencies;

[0043] Calculate the second coefficient of variation according to the newly added clutter frequencies.

[0044] Furthermore, the formula for calculating the first coefficient of variation is: , where μ1 is the first coefficient of variation, f z,i is the i-th newly added important frequency, f z,avg is the frequency mean of the important frequency set of the previous module among adjacent modules, i is a positive integer, N z is the number of newly added important frequencies, and D is the denominator parameter;

[0045] The formula for calculating the second coefficient of variation is: , where μ2 is the second coefficient of variation, f s,i is the i-th newly added clutter frequency, f s,avg is the frequency mean of the clutter frequency set of the previous module among adjacent modules, N s is the number of newly added clutter frequencies.

[0046] Furthermore, the waveform symmetry difference between adjacent modules is equal to the difference between the waveform symmetry value of the previous module and the waveform symmetry value of the subsequent module among adjacent modules;

[0047] The waveform stability difference between adjacent modules is equal to the difference between the waveform symmetry stability value of the previous module and the waveform symmetry stability value of the subsequent module among adjacent modules.

[0048] Furthermore, the convolutional neural network includes: a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a first Concat layer, a second Concat layer, a first fully connected layer, a second fully connected layer, and a weighting layer;

[0049] The input end of the first convolutional layer is used to input the first sample, the input end of the second convolutional layer is used to input the first sample, the input end of the third convolutional layer is used to input the second sample, and the input end of the fourth convolutional layer is used to input the second sample. Among them, the positive half-wave frequency variation value, negative half-wave frequency variation value, waveform symmetry difference, and waveform stability difference of the signal generation module and the operational amplifier module together constitute the first sample, and the positive half-wave frequency variation value, negative half-wave frequency variation value, waveform symmetry difference, and waveform stability difference of the operational amplifier module and the coupling circuit module together constitute the second sample;

[0050] The input ends of the first Concat layer are respectively connected to the output ends of the first convolutional layer and the third convolutional layer; the input ends of the second Concat layer are respectively connected to the output ends of the second convolutional layer and the fourth convolutional layer; the input end of the first fully connected layer is connected to the output end of the first Concat layer; the input end of the second fully connected layer is connected to the output end of the second Concat layer; the input ends of the weighting layer are respectively connected to the output ends of the first fully connected layer and the second fully connected layer, and its output end is used to output the driving chip fault value.

[0051] The beneficial effects of the present invention are as follows: The present invention captures the subtle features of the signal through zero-crossing division and half-wave analysis, then deeply excavates the signal features by using waveform symmetry analysis and frequency domain conversion, and highlights the abnormal features through the comparative analysis of adjacent modules. Finally, a convolutional neural network is used to process the positive half-wave frequency variation value, negative half-wave frequency variation value, waveform symmetry difference, and waveform stability difference of adjacent modules to obtain the driving chip fault value. By dividing the output signal of the present invention into a positive half-wave and a negative half-wave, the fault features are highlighted from the symmetry of the waveform and the frequency domain features between adjacent modules, making the fault features significant and easy to identify, and improving the fault detection accuracy. Description of the Drawings

[0052] Figure 1 is a flowchart of a method for detecting faults in a driving chip of an optoelectronic accelerometer;

[0053] Figure 2 is a schematic structural diagram of the driving chip;

[0054] Figure 3 is a schematic structural diagram of the convolutional neural network. Detailed Embodiments

[0055] The following describes the detailed embodiments of the present invention to facilitate those skilled in the art of the present technology to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the detailed embodiments. For those of ordinary skill in the art of the present technology, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions created using the concept of the present invention are within the scope of protection.

[0056] As Figure 1 shown, a method for detecting faults in a driving chip of an optoelectronic accelerometer includes:

[0057] Dividing the output signals of each module in the driving chip according to zero-crossing points to obtain a positive half-wave set and a negative half-wave set;

[0058] For the positive half-wave set and negative half-wave set of each module, extracting the waveform symmetry value and the waveform symmetry stability value;

[0059] Convert the positive half-wave set and the negative half-wave set to the frequency domain, divide the frequency domain information into important frequencies and clutter frequencies, and obtain four frequency sets. Among them, the four frequency sets include: the positive half-wave important frequency set, the positive half-wave clutter frequency set, the negative half-wave important frequency set, and the negative half-wave clutter frequency set;

[0060] According to the differences between adjacent modules in the driver chip on the four frequency sets, obtain the positive half-wave frequency variation value and the negative half-wave frequency variation value of adjacent modules;

[0061] According to the differences between adjacent modules in the driver chip on the waveform symmetry value and the waveform symmetry stability value, obtain the waveform symmetry difference and the waveform stability difference of adjacent modules;

[0062] Take the positive half-wave frequency variation value, the negative half-wave frequency variation value, the waveform symmetry difference, and the waveform stability difference of each adjacent module as samples and input them into the convolutional neural network to obtain the driver chip fault value.

[0063] In the present invention, the points where the output signals of each module in the driver chip are 0 are marked to obtain the zero-crossing points. According to the positions of the zero-crossing points, the output signals of each module are segmented. The part where the signal segment is greater than 0 constitutes the positive half-wave set, and the part where the signal segment is less than 0 constitutes the negative half-wave set.

[0064] As Figure 2 shown, each module in the driver chip sequentially includes: a signal generation module, an operational amplifier module, and a coupling circuit module; in the present invention, the signal generation module and the operational amplifier module are adjacent modules, the signal generation module is the front module, and the operational amplifier module is the rear module; the operational amplifier module and the coupling circuit module are adjacent modules, the operational amplifier module is the front module, and the coupling circuit module is the rear module.

[0065] In this embodiment, the process of extracting the waveform symmetry value is as follows:

[0066] For each positive half-wave in the positive half-wave set, calculate the positive half-wave area;

[0067] For each negative half-wave in the negative half-wave set, calculate the negative half-wave area;

[0068] Take the positive half-wave area and the peak value of the positive half-wave as elements to form a positive half-wave feature vector;

[0069] Take the negative half-wave area and the absolute value of the peak value of the negative half-wave as elements to form a negative half-wave feature vector;

[0070] Calculate the similarity between the positive half-wave feature vector and the negative half-wave feature vector in the same period;

[0071] Take the average value of the similarities in each period to obtain the waveform symmetry value.

[0072] In a normally operating optoelectronic accelerometer drive chip, the energy carried by the positive and negative half-waves of the signal is relatively stable and has a certain symmetry. For example, in an ideal sine wave signal, the energy of the positive and negative half-waves is equal. When a chip fails, such as a change in gain or non-linear distortion, it will cause an imbalance in the energy distribution between the positive and negative half-waves. By calculating the area, this change in energy distribution can be quantified and used as an important feature for fault detection.

[0073] The positive half-wave feature vector and the negative half-wave feature vector contain two important signal features: area and peak value. Calculating their similarity is actually measuring the consistency of the energy distribution and amplitude characteristics between the positive and negative half-waves. The higher the similarity, the more symmetric the positive and negative half-waves are; conversely, it indicates that the signal is asymmetric, which may be caused by chip failure.

[0074] In this embodiment, the formula for calculating the waveform symmetry stability value is: , where γ is the waveform symmetry stability value, and σ is the standard deviation of the similarity of each period.

[0075] The standard deviation σ is an indicator for measuring the degree of data dispersion. The standard deviation of the similarity of each period can reflect the fluctuation of the waveform symmetry degree in different periods. The larger the standard deviation σ, the smaller the waveform symmetry stability value.

[0076] In this embodiment, , , where A + is the area of the positive half-wave, A - is the area of the negative half-wave, f + (t) is the positive half-wave, f - (t) is the negative half-wave, d is the differential symbol, t is time, | | is the absolute value, T + is the length of the positive half-wave, T - is the length of the negative half-wave.

[0077] In this embodiment, each positive half-wave in the positive half-wave set and each negative half-wave in the negative half-wave set are respectively subjected to time-frequency conversion to obtain positive half-wave frequency domain information and negative half-wave frequency domain information. Each frequency domain information is divided into an important frequency set and a clutter frequency set to obtain a positive half-wave important frequency set, a positive half-wave clutter frequency set, a negative half-wave important frequency set, and a negative half-wave clutter frequency set.

[0078] In this embodiment, the process of obtaining the positive half-wave important frequency set and the positive half-wave clutter frequency set is as follows:

[0079] Each positive half-wave in the positive half-wave set is respectively subjected to time-frequency conversion to obtain positive half-wave frequency domain information;

[0080] Take the amplitudes greater than the amplitude threshold in the positive half-wave frequency domain information as the important amplitudes of the positive half-wave, and take the other amplitudes as the clutter amplitudes of the positive half-wave;

[0081] Arrange the frequencies corresponding to the important amplitudes of the positive half-wave for all periods in sequence to construct a set of important frequencies of the positive half-wave;

[0082] Arrange the frequencies corresponding to the clutter amplitudes of the positive half-wave for all periods in sequence to construct a set of clutter frequencies of the positive half-wave;

[0083] Alternatively, the process of obtaining the set of important frequencies of the negative half-wave and the set of clutter frequencies of the negative half-wave is as follows:

[0084] Perform time-frequency conversion on each negative half-wave in the negative half-wave set respectively to obtain the negative half-wave frequency domain information, and take the absolute value of the amplitudes in the negative half-wave frequency domain information;

[0085] Take the amplitudes greater than the amplitude threshold in the negative half-wave frequency domain information as the important amplitudes of the negative half-wave, and take the other amplitudes as the clutter amplitudes of the negative half-wave;

[0086] Arrange the frequencies corresponding to the important amplitudes of the negative half-wave for all periods in sequence to construct a set of important frequencies of the negative half-wave;

[0087] Arrange the frequencies corresponding to the clutter amplitudes of the negative half-wave for all periods in sequence to construct a set of clutter frequencies of the negative half-wave.

[0088] In the present invention, the amplitude threshold can be set to 0.707 times the maximum value (that is, at -3 dB). Find the maximum amplitude from the positive half-wave frequency domain information, and the corresponding amplitude threshold is 0.707 times the maximum amplitude; find the maximum amplitude from the negative half-wave frequency domain information after taking the absolute value, and the corresponding amplitude threshold is 0.707 times the maximum amplitude.

[0089] In the actual operation of the optoelectronic accelerometer drive chip, when the chip is operating normally, the frequency components of the output signal have a certain stability. When the chip fails, such as circuit parameter changes, component damage, etc., it will cause abnormal changes in the amplitudes of some frequency components in the signal. By obtaining the set of important frequencies and the set of clutter frequencies, these abnormal frequency characteristics can be highlighted.

[0090] In this embodiment, the process of obtaining the positive half-wave frequency variation value and the negative half-wave frequency variation value of adjacent modules includes:

[0091] Obtain the first variation coefficient of the positive half-wave according to the difference between adjacent modules in the set of important frequencies of the positive half-wave in the drive chip; obtain the first variation coefficient of the negative half-wave according to the difference between adjacent modules in the set of important frequencies of the negative half-wave in the drive chip;

[0092] Obtain the second coefficient of variation for the positive half-wave based on the difference in the clutter frequency sets of adjacent modules in the driving chip; obtain the second coefficient of variation for the negative half-wave based on the difference in the clutter frequency sets of adjacent modules in the driving chip.

[0093] Weight the first coefficient of variation for the positive half-wave and the second coefficient of variation for the positive half-wave to obtain the frequency variation value of adjacent modules for the positive half-wave.

[0094] Weight the first coefficient of variation for the negative half-wave and the second coefficient of variation for the negative half-wave to obtain the frequency variation value of adjacent modules for the negative half-wave.

[0095] In this embodiment, when weighting the first coefficient of variation and the second coefficient of variation, the first coefficient of variation is mainly considered. Therefore, a larger weight value is set for the first coefficient of variation. For example, the weight value of the first coefficient of variation can be set to 0.6, and the weight value of the second coefficient of variation can be set to 0.4.

[0096] In this embodiment, the process of obtaining the first coefficient of variation includes:

[0097] Take the intersection of the important frequency sets of adjacent modules to obtain the important frequency intersection.

[0098] Exclude the important frequency intersection from the important frequency set of the subsequent module among adjacent modules to obtain the newly added important frequencies.

[0099] Calculate the first coefficient of variation based on the newly added important frequencies.

[0100] Furthermore, the process of obtaining the first coefficient of variation for the positive half-wave includes:

[0101] Take the intersection of the positive half-wave important frequency sets of adjacent modules to obtain the positive half-wave important frequency intersection.

[0102] Exclude the positive half-wave important frequency intersection from the positive half-wave important frequency set of the subsequent module among adjacent modules to obtain the newly added positive half-wave important frequencies.

[0103] Calculate the first coefficient of variation for the positive half-wave based on the newly added positive half-wave important frequencies.

[0104] Furthermore, the process of obtaining the first coefficient of variation for the negative half-wave includes:

[0105] Take the intersection of the negative half-wave important frequency sets of adjacent modules to obtain the negative half-wave important frequency intersection.

[0106] Exclude the negative half-wave important frequency intersection from the negative half-wave important frequency set of the subsequent module among adjacent modules to obtain the newly added negative half-wave important frequencies.

[0107] Calculate the first coefficient of variation for the negative half-wave based on the newly added negative half-wave important frequencies.

[0108] In this embodiment, the process of obtaining the second coefficient of variation includes:

[0109] Taking the intersection of the clutter frequency sets of adjacent modules to obtain a clutter frequency intersection;

[0110] Removing the clutter frequency intersection from the clutter frequency set of the subsequent module among adjacent modules to obtain new clutter frequencies;

[0111] Calculating the second coefficient of variation according to the new clutter frequencies.

[0112] Further, the process of obtaining the second coefficient of variation of the positive half-wave includes:

[0113] Taking the intersection of the positive half-wave clutter frequency sets of adjacent modules to obtain a positive half-wave clutter frequency intersection;

[0114] Removing the positive half-wave clutter frequency intersection from the positive half-wave clutter frequency set of the subsequent module among adjacent modules to obtain new positive half-wave clutter frequencies;

[0115] Calculating the second coefficient of variation of the positive half-wave according to the new positive half-wave clutter frequencies.

[0116] Further, the process of obtaining the second coefficient of variation of the negative half-wave includes:

[0117] Taking the intersection of the negative half-wave clutter frequency sets of adjacent modules to obtain a negative half-wave clutter frequency intersection;

[0118] Removing the negative half-wave clutter frequency intersection from the negative half-wave clutter frequency set of the subsequent module among adjacent modules to obtain new negative half-wave clutter frequencies;

[0119] Calculating the second coefficient of variation of the negative half-wave according to the new negative half-wave clutter frequencies.

[0120] Dividing the frequency set into important frequencies and clutter frequencies, and calculating the corresponding coefficients of variation respectively, which follows the principle of distinguishing key signal components and interference components in signal analysis. The important frequency set reflects the key frequency characteristics when the chip is working normally, while the clutter frequency set contains interference or abnormal frequency components. Calculating the coefficients of variation separately and weighting them enables considering both the changes in key frequencies and the influence of interference frequencies when analyzing frequency variation, and can more accurately reflect the changes in the chip working state.

[0121] In this embodiment, the formula for calculating the first coefficient of variation is: , where μ1 is the first coefficient of variation, f z,i is the i-th new important frequency, f z,avg is the frequency mean of the important frequency set of the previous module among adjacent modules, i is a positive integer, Nz is the number of newly added important frequencies, and D is the denominator parameter.

[0122] In this embodiment, the formula for calculating the second coefficient of variation is: , where μ2 is the second coefficient of variation, and f s,i is the i-th newly added clutter frequency, and f s,avg is the frequency mean of the clutter frequency set of the previous module in the adjacent module, and N s is the number of newly added clutter frequencies.

[0123] The present invention quantifies the difference degree of important frequencies or clutter frequencies between adjacent modules by calculating the sum of squared deviations of the newly added frequencies from the frequency mean of the previous module. The denominator parameter D is a parameter greater than 0, which is used to prevent the denominator from being 0. At the same time, the value ranges of the first coefficient of variation and the second coefficient of variation are between 0 and 1.

[0124] In this embodiment, the waveform symmetry difference between adjacent modules is equal to the difference between the waveform symmetry value of the previous module in the adjacent modules and the waveform symmetry value of the subsequent module in the adjacent modules;

[0125] The waveform stability difference between adjacent modules is equal to the difference between the waveform symmetry stability value of the previous module in the adjacent modules and the waveform symmetry stability value of the subsequent module in the adjacent modules.

[0126] The present invention can intuitively reflect the change trend of the signal waveform symmetry characteristics and their stability in the process of transmission between modules by calculating the differences between the waveform symmetry values and waveform symmetry stability values of adjacent modules. When each module of the chip works normally, each module maintains stability in the transmission of signals. If a large difference appears, it indicates that abnormal changes have occurred in the waveform symmetry characteristics and stability during the transmission of signals between modules.

[0127] As Figure 3 shown, the convolutional neural network includes: a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a first Concat layer, a second Concat layer, a first fully connected layer, a second fully connected layer, and a weighting layer;

[0128] The input end of the first convolutional layer is used to input the first sample, the input end of the second convolutional layer is used to input the first sample, the input end of the third convolutional layer is used to input the second sample, and the input end of the fourth convolutional layer is used to input the second sample. Among them, the positive half-wave frequency variation value, negative half-wave frequency variation value, waveform symmetry difference, and waveform stability difference of the signal generation module and the operational amplifier module together constitute the first sample, and the positive half-wave frequency variation value, negative half-wave frequency variation value, waveform symmetry difference, and waveform stability difference of the operational amplifier module and the coupling circuit module together constitute the second sample;

[0129] The input ends of the first Concat layer are respectively connected to the output ends of the first convolutional layer and the third convolutional layer; the input ends of the second Concat layer are respectively connected to the output ends of the second convolutional layer and the fourth convolutional layer; the input end of the first fully connected layer is connected to the output end of the first Concat layer; the input end of the second fully connected layer is connected to the output end of the second Concat layer; the input ends of the weighting layer are respectively connected to the output ends of the first fully connected layer and the second fully connected layer, and its output end is used to output the driving chip fault value.

[0130] In this embodiment, the first convolutional layer and the third convolutional layer adopt convolutional layers with a convolutional kernel of in size, the second convolutional layer and the fourth convolutional layer adopt convolutional layers with a convolutional kernel of in size, the first sample is a matrix formed by arranging 4 values into and the second sample is a matrix formed by arranging 4 values into .

[0131] The present invention respectively uses convolutional layers with a size of to extract features from the first sample and the second sample. The first Concat layer splices the extracted features and uses the first fully connected layer to calculate the first fault component; for the first sample and the second sample, convolutional layers with a size of are used to extract features again. The second Concat layer splices the extracted features and uses the second fully connected layer to calculate the second fault component. The first fault component and the second fault component are weighted in the weighting layer to obtain the driving chip fault value.

[0132] The present invention respectively uses convolutional layers with sizes of and to extract features from the first sample and the second sample, and can capture the feature information of the samples from different scales. The first Concat layer splices the features at one scale, and the second Concat layer splices the features at another scale to realize dual-channel calculation of the fault component and improve the accuracy of fault detection.

[0133] In this embodiment, the weighting layer can be implemented by a fully connected layer.

[0134] In this embodiment, the convolutional neural network can be trained by the existing gradient descent method. A data set containing the first sample and the second sample is collected, and at the same time, the true driving chip fault value (label) corresponding to each sample is recorded. The data set is input into the convolutional neural network, and the mean square error (MSE) loss function is selected for training.

[0135] The present invention captures the subtle features of a signal through zero-crossing division and half-wave analysis, then deeply excavates the signal features by using waveform symmetry analysis and frequency-domain conversion, and highlights the abnormal features through comparative analysis of adjacent modules. Finally, a convolutional neural network is used to process the positive half-wave frequency variation value, negative half-wave frequency variation value, waveform symmetry difference value, and waveform stability difference value of adjacent modules to obtain the fault value of the drive chip. By dividing the output signal into a positive half-wave and a negative half-wave, the present invention highlights the fault features from the symmetry of the waveform and the frequency-domain features between adjacent modules, making the fault features significant and easy to identify, and improving the fault detection accuracy.

[0136] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for detecting faults in a photoelectric accelerometer driver chip, characterized in that: include: The output signals of each module in the driver chip are divided according to the zero crossing point to obtain a positive half-wave set and a negative half-wave set; For the positive half-wave set and the negative half-wave set of each module, extract the waveform symmetry value and the waveform symmetry stability value; The positive half-wave set and the negative half-wave set are converted into the frequency domain, and the frequency domain information is divided into important frequencies and clutter frequencies, so as to obtain four frequency sets, wherein the four frequency sets include: a positive half-wave important frequency set, a positive half-wave clutter frequency set, a negative half-wave important frequency set and a negative half-wave clutter frequency set; According to the gaps between adjacent modules in the driver chip in the four frequency sets, the positive half-wave frequency variation value and the negative half-wave frequency variation value of the adjacent modules are obtained; According to the differences in waveform symmetry values ​​and waveform symmetry stability values ​​of adjacent modules in the driver chip, the waveform symmetry difference and waveform stability difference of adjacent modules are obtained; The positive half-wave frequency variation value, negative half-wave frequency variation value, waveform symmetry difference value and waveform stability difference value of each adjacent module are input into the convolutional neural network as samples to obtain the fault value of the driver chip; The process of extracting the waveform symmetry value is: For each positive half-wave in the positive half-wave set, calculate the positive half-wave area; For each negative half-wave in the negative half-wave set, calculate the negative half-wave area; The positive half-wave area and the positive half-wave peak value are used as elements to form a positive half-wave eigenvector; The negative half-wave area and the absolute value of the peak value of the negative half-wave are used as elements to form a negative half-wave eigenvector; Calculate the similarity between the positive half-wave eigenvector and the negative half-wave eigenvector of the same period; Take the average of the similarities of each cycle to obtain the waveform symmetry value; The formula for calculating the waveform symmetry stability value is: , where γ is the waveform symmetry stability value and σ is the standard deviation of the similarity of each cycle.

2. The photoelectric accelerometer driver chip fault detection method according to claim 1, characterized in that: Each positive half-wave in the positive half-wave set and each negative half-wave in the negative half-wave set is converted from time to frequency to obtain positive half-wave frequency domain information and negative half-wave frequency domain information, and each frequency domain information is divided into an important frequency set and a clutter frequency set to obtain a positive half-wave important frequency set, a positive half-wave clutter frequency set, a negative half-wave important frequency set and a negative half-wave clutter frequency set.

3. The photoelectric accelerometer driver chip fault detection method according to claim 1, characterized in that: The process of obtaining the positive half-wave important frequency set and the positive half-wave clutter frequency set is: Perform time-frequency conversion on each positive half-wave in the positive half-wave set to obtain frequency domain information of the positive half-wave; The amplitude greater than the amplitude threshold in the frequency domain information of the positive half-wave is regarded as the important amplitude of the positive half-wave, and the other amplitudes are regarded as the positive half-wave clutter amplitude; The frequencies corresponding to the important amplitudes of the positive half-waves of all cycles are arranged in sequence to construct a set of important frequencies of the positive half-waves; The frequencies corresponding to the positive half-wave clutter amplitudes of all cycles are arranged in sequence to construct a positive half-wave clutter frequency set; Alternatively, the process of obtaining the negative half-wave important frequency set and the negative half-wave clutter frequency set is: Performing time-frequency conversion on each negative half-wave in the negative half-wave set respectively to obtain frequency domain information of the negative half-wave, and taking the absolute value of the amplitude in the frequency domain information of the negative half-wave; The amplitude greater than the amplitude threshold in the frequency domain information of the negative half-wave is regarded as the important amplitude of the negative half-wave, and the other amplitudes are regarded as the clutter amplitude of the negative half-wave; The frequencies corresponding to the important amplitudes of the negative half-waves of all cycles are arranged in sequence to construct a set of important frequencies of the negative half-waves; The frequencies corresponding to the negative half-wave clutter amplitudes of all cycles are arranged in sequence to construct a negative half-wave clutter frequency set.

4. The photoelectric accelerometer driver chip fault detection method according to claim 1, characterized in that: The process of obtaining the positive half-wave frequency variation value of the adjacent module and the negative half-wave frequency variation value of the adjacent module includes: According to the difference between adjacent modules in the driver chip in the positive half-wave important frequency set, the first variation coefficient of the positive half-wave is obtained; according to the difference between adjacent modules in the driver chip in the negative half-wave important frequency set, the first variation coefficient of the negative half-wave is obtained; According to the difference between the adjacent modules in the driver chip in the positive half-wave noise frequency set, the second variation coefficient of the positive half-wave is obtained; according to the difference between the adjacent modules in the driver chip in the negative half-wave noise frequency set, the second variation coefficient of the negative half-wave is obtained; The first variation coefficient of the positive half-wave and the second variation coefficient of the positive half-wave are weighted to obtain the frequency variation value of the positive half-wave of the adjacent modules; The first variation coefficient of the negative half-wave and the second variation coefficient of the negative half-wave are weighted to obtain the frequency variation value of the negative half-wave of the adjacent modules.

5. The photoelectric accelerometer driver chip fault detection method according to claim 4, characterized in that: The process of obtaining the first coefficient of variation includes: Take the intersection of the important frequency sets of adjacent modules to obtain the important frequency intersection; Eliminate the intersection of important frequencies from the important frequency set of the subsequent module in the adjacent module to obtain the newly added important frequency; Calculate the first coefficient of variation based on the newly added important frequencies; Alternatively, the process of obtaining the second coefficient of variation includes: Taking the intersection of the clutter frequency sets of adjacent modules to obtain the clutter frequency intersection; Eliminate the intersection of clutter frequencies from the clutter frequency set of the subsequent module in the adjacent module to obtain the newly added clutter frequency; The second coefficient of variation is calculated based on the newly added clutter frequency.

6. The photoelectric accelerometer driver chip fault detection method according to claim 5, characterized in that: The formula for calculating the first coefficient of variation is: , where μ1 is the first coefficient of variation, f z,i is the i-th newly added important frequency, f z,avg is the frequency mean of the important frequency set of the previous module in the adjacent module, i is a positive integer, N z is the number of newly added important frequencies, and D is the denominator parameter; The formula for calculating the second coefficient of variation is: , where μ2 is the second coefficient of variation, f s,i is the ith new clutter frequency, f s,avg is the frequency mean of the clutter frequency set of the previous module in the adjacent module, N s is the number of newly added clutter frequencies.

7. The photoelectric accelerometer driver chip fault detection method according to claim 1, characterized in that: The waveform symmetry difference of adjacent modules is equal to the difference between the waveform symmetry value of the preceding module in the adjacent modules and the waveform symmetry value of the following module in the adjacent modules; The waveform stability difference between adjacent modules is equal to the difference between the waveform symmetry stability value of the preceding module in the adjacent modules and the waveform symmetry stability value of the following module in the adjacent modules.

8. The photoelectric accelerometer driver chip fault detection method according to claim 1, characterized in that: The convolutional neural network includes: a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a first Concat layer, a second Concat layer, a first fully connected layer, a second fully connected layer and a weighted layer; The input end of the first convolution layer is used to input the first sample, the input end of the second convolution layer is used to input the first sample, the input end of the third convolution layer is used to input the second sample, and the input end of the fourth convolution layer is used to input the second sample, wherein the positive half-wave frequency variation value, the negative half-wave frequency variation value, the waveform symmetry difference value and the waveform stability difference value of the signal generation module and the operational amplifier module together constitute the first sample, and the positive half-wave frequency variation value, the negative half-wave frequency variation value, the waveform symmetry difference value and the waveform stability difference value of the operational amplifier module and the coupling circuit module together constitute the second sample; The input end of the first Concat layer is connected to the output end of the first convolutional layer and the output end of the third convolutional layer respectively; the input end of the second Concat layer is connected to the output end of the second convolutional layer and the output end of the fourth convolutional layer respectively; the input end of the first fully connected layer is connected to the output end of the first Concat layer; the input end of the second fully connected layer is connected to the output end of the second Concat layer; the input end of the weighted layer is connected to the output end of the first fully connected layer and the output end of the second fully connected layer respectively, and its output end is used to output the fault value of the driver chip.

Images