CN103824135B - A kind of analog circuit fault Forecasting Methodology - Google Patents
A kind of analog circuit fault Forecasting Methodology Download PDFInfo
- Publication number
- CN103824135B CN103824135B CN201410088347.6A CN201410088347A CN103824135B CN 103824135 B CN103824135 B CN 103824135B CN 201410088347 A CN201410088347 A CN 201410088347A CN 103824135 B CN103824135 B CN 103824135B
- Authority
- CN
- China
- Prior art keywords
- fault
- analog circuit
- prediction
- failure
- health
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 36
- 239000013598 vector Substances 0.000 claims abstract description 78
- 230000036541 health Effects 0.000 claims abstract description 44
- 238000012360 testing method Methods 0.000 claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 9
- 238000004458 analytical method Methods 0.000 claims abstract description 6
- 239000002245 particle Substances 0.000 claims description 45
- 238000003745 diagnosis Methods 0.000 claims description 16
- 230000004044 response Effects 0.000 claims description 15
- 238000005457 optimization Methods 0.000 claims description 13
- 238000013528 artificial neural network Methods 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 5
- 230000009466 transformation Effects 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 abstract description 2
- 230000007935 neutral effect Effects 0.000 abstract 1
- 238000004364 calculation method Methods 0.000 description 5
- 238000000354 decomposition reaction Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 239000003990 capacitor Substances 0.000 description 3
- 230000032683 aging Effects 0.000 description 2
- 230000015556 catabolic process Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000006731 degradation reaction Methods 0.000 description 2
- 230000008030 elimination Effects 0.000 description 2
- 238000003379 elimination reaction Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000004836 empirical method Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- BULVZWIRKLYCBC-UHFFFAOYSA-N phorate Chemical compound CCOP(=S)(OCC)SCSCC BULVZWIRKLYCBC-UHFFFAOYSA-N 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Abstract
A kind of analog circuit fault Forecasting Methodology, comprises the following steps: element each to analog circuit carries out Monte Carlo Analysis in fault-free interval and extract each band signal energy；The band signal energy normalized that will extract, obtains characteristic vector；Training BP neutral net；Judge the failure mode having occurrence tendency；Extraction element is positioned at failure predication characteristic vector during initial value；Extract failure predication characteristic vector during circuit-under-test work；Calculate cosine angle distance and characterize the health degree of element；Health degree threshold value when computing element breaks down；The kernel function width factor of Method Using Relevance Vector Machine algorithm is in optimized selection；Analog circuit is carried out failure predication.The present invention both may be used for real-time system, it is also possible to for non real-time system；Both the fault of In Linear Analog Circuits can be predicted, it is also possible to the fault of non-linear analog circuit is predicted；The main elements such as resistance, inductance and electric capacity in analog circuit can be carried out failure predication.
Description
Technical Field
The invention relates to a method for predicting a fault of an analog circuit, in particular to a method for predicting the fault of the analog circuit by establishing a fault prediction model.
Background
The analog circuit is widely applied to household appliances, industrial production lines, automobiles, aerospace and other equipment, and the performance reduction, the function failure, the slow response and other electronic faults of the equipment can be caused by the faults of the analog circuit. It is therefore necessary to evaluate the state of the analog circuit.
The state evaluation of analog circuits generally includes fault diagnosis and fault prediction. The fault diagnosis is developed quickly, and the accuracy of fault diagnosis in a large amount of research works can reach about 99%. Current analog circuit fault prediction efforts are generally directed to specific components of the analog circuit, rather than to the circuit as a whole. One difficulty with not being able to predict the circuit as a whole is that there are few ways to accurately describe the performance degradation, i.e. the health degradation, of the individual elements of an analog circuit. Meanwhile, the fault of the nonlinear analog circuit can be predicted by the current method.
The correlation vector machine is a regression prediction algorithm based on a Bayesian frame, has high operation speed, is suitable for online detection, and has been proved by research to have higher prediction precision than common algorithms such as a support vector machine, a neural network and the like. The width factor of the kernel function in the correlation vector machine algorithm has great influence on the prediction precision, and the width factor is obtained by adopting an empirical method in the past.
Disclosure of Invention
The invention aims to solve the technical problem of overcoming the defects in the prior art and provides an analog circuit fault prediction method with high fault prediction precision. The method comprises the steps of firstly identifying an element with a failure tendency of a circuit and a failure occurrence mode (a forward deviation nominal value or a reverse deviation nominal value) of the element through a BP neural network, obtaining data of the change of the health degree of the element along with time points by calculating cosine similarity of different time points, and then predicting whether the element fails at a certain future time point or not through a related vector machine algorithm optimized based on a particle swarm algorithm, or directly predicting the failure occurrence time point of the element. The method is popular and effective for main elements such as resistors, capacitors and inductors in an analog circuit.
The technical scheme adopted by the invention for solving the technical problems is as follows:
a method for predicting the failure of an analog circuit comprises the following steps:
(1) carrying out Monte Carlo analysis on each element of the analog circuit in a fault-free interval, extracting a test node signal, carrying out wavelet packet conversion noise elimination processing on the extracted signal, and extracting signal energy of each frequency band, wherein the test node signal is generally branch voltage;
(2) normalizing the extracted frequency band signal energy to obtain a fault diagnosis characteristic vector;
(3) taking the fault diagnosis feature vector as training data to train a BP neural network;
(4) extracting node signals of a circuit to be detected during working, carrying out wavelet packet transformation and normalization to generate corresponding fault diagnosis characteristic vectors, and judging the types of faults with trends by using a BP neural network;
(5) extracting a frequency sweep response signal of a tested circuit test node when an element is positioned at an initial value so as to form a fault prediction characteristic vector of the initial value, wherein the response signal is generally branch voltage;
(6) extracting a frequency sweep response signal of a test node of the analog circuit to be tested during working according to a fixed time interval so as to form a fault prediction characteristic vector of a time point sequence of an element;
(7) calculating the cosine angle distance between the element fault prediction characteristic vector extracted according to the fixed time interval and the fault prediction characteristic vector of the element initial value, and representing the health degree of the element at different time points to generate a corresponding health degree sequence;
(8) calculating a health threshold when the element fails;
(9) based on the element health degree sequence and time point sequence data, optimizing and selecting a kernel function width factor of a correlation vector machine algorithm by applying a particle swarm algorithm;
(10) and (3) performing fault prediction on the analog circuit by using a correlation vector machine algorithm optimized by a particle swarm algorithm.
In the step (1), the non-fault section is [50%,1+ lower limit of normal tolerance ] of the nominal value of the analog circuit element and [1+ upper limit of normal tolerance, 150% ] of the nominal value. The Monte Carlo analysis of each element of the analog circuit in the fault-free interval can adopt the Monte Carlo function of Pspice software.
In the step (4), the failure type that tends to occur refers to the element that has failed and the direction in which the element deviates from the initial value when the element fails.
In the step (5), the initial value is defined as: the initial value is equal to the nominal value x (1 + upper normal tolerance limit) when the component fails in a forward direction deviating from the nominal value, and is equal to the nominal value x (1 + lower normal tolerance limit) when the component fails in a reverse direction deviating from the nominal value. And the alternating current scanning function of Pspice software can be adopted for extracting the scanning response signal of the initial value.
In the step (6), the fixed interval time for extracting the test node sweep frequency response signal of the tested analog circuit is sequentiallyT 1 , T 2 ,…, T n The time point sequence is [ [ 2 ] ]T 1 , T 2 ,…, T n ]，Is the total number of time points.
In the step (7), the cosine angular distance method includes:
；
wherein,characterized by a fault prediction feature vector when the element is at an initial value,representing a vectorTo middlejThe characteristics of the device are as follows,，mis the number of features in the extracted set of feature vectors;characterized by the firstiThe component failure prediction feature vectors extracted at each time point,representing a vectorTo middlejThe characteristics of the device are as follows,，is the total number of extracted feature quantities, and is also the total number of time points.
Degree of healthThe health degree sequence generated by the element failure prediction feature vector extracted at regular time intervals is [ alpha ]H 1 , H 2 ,…, H n ]The corresponding time point sequence is [ 2 ]T 1 , T 2 ,…, T n ]。
In the step (8), the method for calculating the health threshold when the element fails includes: and calculating the cosine angular distance of the final deviation value in the fault-free interval of the deviation direction of the element, namely the health threshold value of the element in the deviation direction with fault. The health threshold when a component fails may also be referred to as a health failure threshold for short.
In the step (9), the step of applying the particle swarm algorithm to optimize and select the kernel function width factor of the correlation vector machine algorithm includes:
(1) initializing particle swarm algorithm parameters including position, speed, optimizing range and iteration times, wherein the width factor is mapped to the position of the particle;
(2) calculating the fitness of each particle, and obtaining the individual optimal position of each particle and the global optimal position of the particle swarm according to the fitness;
(3) updating the speed and the position of each particle;
(4) and (5) repeating the steps (2) and (3) until the iteration is finished, and outputting the result.
In the step (10), the failure prediction of the analog circuit includes two modes:
(1) predicting a failure occurrence time point of the element;
(2) and predicting whether the element fails at a certain future time point.
In the invention, the cosine angle distance is adopted to simply and effectively represent the health degree of elements in the analog circuit. When any main element in the circuit gradually deviates from the initial value, the calculated cosine angular distance is reduced, and the health degree of the characteristic element is reduced, which is equivalent to the reduction of the health degree of the circuit. The particle swarm algorithm is adopted to optimize and select the kernel function width factor of the correlation vector machine algorithm, so that the fault prediction precision can be further improved. The invention is simple and effective, and can be used for a real-time system and a non-real-time system; the fault of the linear analog circuit can be predicted, and the fault of the nonlinear analog circuit can also be predicted; the fault prediction can be carried out on main elements such as resistance, inductance and capacitance in the analog circuit.
Compared with the background technology, the invention has the following advantages:
(1) the fault prediction characteristic vector is simple to extract and is suitable for a real-time environment;
(2) the cosine angular distance can be directly calculated by applying the extracted fault prediction characteristic vector, preprocessing is not needed, the calculation is simple, and the method is also suitable for a real-time environment;
(3) the method can be applied to the fault prediction of the nonlinear analog circuit as well as the branch voltage is extracted as the fault prediction characteristic vector;
(4) a method for calculating the cosine angle distance to represent the health degree of each main element of the analog circuit is provided;
(5) after the kernel function width factor is optimized by the particle swarm optimization, the prediction performance of the correlation vector machine algorithm is obviously improved.
Drawings
FIG. 1 is a block flow diagram of a method for analog circuit fault prediction in accordance with the present invention;
FIG. 2 is a block diagram of a process for optimizing a correlation vector machine algorithm by a particle swarm algorithm;
FIG. 3 is a schematic diagram of a Sallen-Key band-pass filter circuit;
FIG. 4 is a graph of the health of R2 versus time points;
FIG. 5 is a diagram of the prediction effect of the particle swarm optimization-based correlation vector machine algorithm.
Detailed Description
The invention is further illustrated by the following figures and examples.
Referring to fig. 1, the overall flow chart of the present invention consists of 8 steps:
step 1: carrying out Monte Carlo analysis on each element of the analog circuit in a fault-free interval, extracting a test node signal, carrying out wavelet packet conversion noise elimination processing on the extracted signal, and extracting signal energy of each frequency band, wherein the test node signal is generally branch voltage; normalizing the extracted frequency band signal energy to obtain a fault diagnosis characteristic vector;
step 2: taking the fault diagnosis feature vector as training data to train a BP neural network;
and step 3: extracting node signals of a circuit to be detected during working, carrying out wavelet packet transformation and normalization to generate corresponding fault diagnosis characteristic vectors, and judging the types of faults with trends by using a BP neural network;
and 4, step 4: extracting the frequency sweep response signals of the tested circuit test node when the fault element is positioned at the initial value according to the fault type determined in the step 3, forming a fault prediction characteristic vector of the initial value, and extracting the frequency sweep response signals of the tested analog circuit test node when the tested analog circuit test node works according to a fixed time interval, forming a fault prediction characteristic vector of a time point sequence of the element;
and 5: calculating the cosine angle distance between the element fault prediction characteristic vector extracted by the element at fixed time intervals and the fault prediction characteristic vector of the initial value of the element, representing the health degree of the element at different time points, and corresponding the element health degree sequence to the time point sequence;
step 6: calculating a health threshold when the element fails;
and 7: based on the health degree sequence and time point sequence data of the elements, optimizing and selecting a kernel function width factor of a correlation vector machine algorithm by applying a particle swarm algorithm;
and 8: and performing fault prediction on the analog circuit by applying the optimized correlation vector machine algorithm.
The 8 steps are divided into two parts in series, namely a fault diagnosis part and a fault prediction part, wherein the fault diagnosis part comprises the steps 1, 2 and 3, and the fault prediction part comprises the steps 4, 5, 6, 7 and 8. The fault diagnosis part aims to identify the fault types with the occurrence tendency, and the fault prediction part establishes a fault prediction model in a targeted manner according to the fault types with the occurrence tendency identified by the fault diagnosis part, selects the corresponding health fault threshold value and performs fault prediction on the analog circuit.
In step 1, wavelet packet transformation denoising processing is performed on the extracted signals, and the method for extracting the energy of each frequency band signal comprises the following steps:
(1) carrying out N layers of orthogonal wavelet packet decomposition on the extracted signal to obtain a high-frequency wavelet packet decomposition coefficient sequence and a low-frequency wavelet packet decomposition coefficient sequence on each scale function space;
(2) denoising the high-frequency coefficient;
(3) and calculating the energy of each layer of wavelet packet decomposition coefficient sequence.
The Monte Carlo analysis of each element of the analog circuit in the fault-free interval can adopt the Monte Carlo function of Pspice software.
The non-fault interval is [50%,1+ lower normal tolerance ] of the nominal value of the element and [1+ upper normal tolerance, 150% ] of the nominal value, when the parameter value of the element deviates from the nominal value by more than 50%, the element can be judged as a fault element, the normal tolerance of the resistance is +/-5%, the normal tolerance of the capacitance is +/-10%, and the non-fault interval represents the working state that an analog circuit element deviates from the normal value but does not reach the fault degree. The normal value of the element is [1+ lower normal tolerance, 1+ upper normal tolerance ] of the nominal value.
The normalization is performed to eliminate the influence of different dimensions of the input data on the calculation.
In step 2, when training the BP neural network, the number of input nodes of the BP neural network is equal to M (N + 1), wherein M is the number of extracted nodes, and N is the number of layers of wavelet packet decomposition; the number of output nodes is equal to the typical number of faults. The hidden layer neuron activation function adopts a Sigmoid function.
In step 3, the output sequence of the BP neural network is appliedAnd if the q number in the sequence is 1, indicating that the circuit will have q faults, and if the other number in the sequence is 0, indicating that other faults do not occur. The trending failure category refers to the failed component and the direction in which the component deviates from the initial value when the failure occurs, i.e., the component deviates from the initial value in the forward direction or in the reverse direction.
In step 4, the initial value is defined as: the initial value is equal to the nominal value x (1 + upper normal tolerance limit) when the component fails in a forward direction deviating from the nominal value, and is equal to the nominal value x (1 + lower normal tolerance limit) when the component fails in a reverse direction deviating from the nominal value. And the alternating current scanning function of Pspice software can be adopted for extracting the scanning response signal of the initial value.
The fixed interval time points for extracting the sweep frequency response signal of the test node of the tested analog circuit are sequentiallyT 1 , T 2 ,…, T n The time point sequence is [ [ 2 ] ]T 1 , T 2 ,…, T n ]Is the total number of time points.
In step 5, the method for calculating the cosine angular distance comprises the following steps:
；
wherein,characterized by a fault prediction feature vector when the element is at an initial value,representing a vectorTo middlejThe characteristics of the device are as follows,，mis the number of features in the extracted set of feature vectors;characterized by the firstiThe component failure prediction feature vectors extracted at each time point,representing a vectorTo middlejThe characteristics of the device are as follows,，is the total number of extracted feature quantities, and is also the total number of time points.
Characterization of component health using cosine angular distanceThe health degree sequence generated by the element failure prediction feature vector extracted at regular time intervals is [ alpha ]H 1 , H 2 ,…, H n ]The corresponding time point sequence is [ 2 ]T 1 , T 2 ,…, T n ]。
When the element is located at the initial value, the cosine angular distance obtained by calculation is 1, the health degree represented by the element is also 1, and when the value of the element deviates from the initial value step by step, the cosine angular distance corresponding to the element is gradually reduced, namely the health degree is reduced.
In step 6, the cosine angular distance of the final deviation value of the non-fault section of the deviation direction of the element is calculated, namely the health degree threshold value of the element in the deviation direction with fault. The health threshold when a component fails may also be referred to as a health failure threshold for short. The final deviation value of the fault-free interval can be expressed as a nominal value × 0.5 and a nominal value × 1.5 in the two deviation directions, respectively.
In step 7, a particle swarm algorithm is applied to optimize and select a proper kernel function width factor for the correlation vector machine algorithm, and the step of optimizing the correlation vector machine algorithm by the particle swarm algorithm can be divided into the following 5 steps with reference to fig. 2:
(1) initializing particle swarm algorithm parameters including position, speed, optimizing range, iteration times and the like, wherein the width factor is mapped to the position of the particle;
(2) calculating the fitness of each particle according to the fitness function, and obtaining the individual optimal position of each particle and the global optimal position of the particle swarm according to the fitness;
(3) updating the speed and the position of each particle;
(4) repeating (2) and (3) until the iteration cycle is finished, namely reaching the maximum number of iterations;
(5) and outputting the optimization result of the particle swarm optimization as a kernel function width factor of the correlation vector machine algorithm.
The calculation formula of the particle swarm algorithm is as follows:
；
；
wherein,is the number of iterations;，the number of particles in the population;is a particleThe position in the seek;is a particleSpeed in the seek;andis an acceleration factor;andis a random number between 0 and 1;is the inertial weight.Is the individual optimal position of the particles in the optimizing process,is the global optimum position of the particle swarm.
The fitness function chosen is:
；
whereinIs the mean square error, which is a minimized objective function that characterizes how closely the predicted data is to the actual data;,is the amount of data used for training;andrespectively prediction data and real data.
In step 8, the steps of the algorithm prediction of the relevance vector machine are as follows:
the prediction output of the correlation vector machine algorithm is:
；
whereinIs a training sample;is a mean of 0 and a variance ofThe noise of (2);is an output function whose expression is:
；
whereinIs the weight of the prediction model;is a deviation;the u training sample is obtained;the gaussian radial basis kernel function, which is a kernel function, consistently has excellent nonlinear processing capability, so that the function is adopted by the people:
；
whereinThe width factor has a crucial effect on the prediction performance of the correlation vector machine algorithm and is optimally selected by the particle swarm optimization.
The fault prediction of the analog circuit comprises two modes:
(1) predicting a failure occurrence time point of the element;
(2) and predicting whether the element fails at a certain future time point.
The prediction method of the mode (1) is as follows:
using optimized correlation vector machine algorithm pairT n Then, the component health degree of each time point is continuously predicted, and the satisfaction is found in the time point sequenceH n+g ≥H threshold And isH n+g-1<HthresholdOf the conditionT n+g At a point in time, the analog circuit being in the futureT n+g A point in time is at fault. WhereinH threshold Is the health threshold of the corresponding fault category when the analog circuit fails.
The prediction method of the mode (2) is as follows:
suppose the predicted element isT n After the time pointkA point in timeT n+k Whether a failure has occurred. Then the optimized relevance vector machine algorithm is used for predictionT n+k Health at time point, assumed to beH n+k 。
① ifH n+k >H threshold Analog circuits in the futurekNo failure occurred at each time point;
② ifH n+k =H threshold Analog circuits in the futurekA fault occurs at each point in time;
③ ifH n+k <H threshold Analog circuits in the futurekAnd (3) faults occur in each time point, and the specific time point of the faults can be judged by referring to the prediction method in the mode (1).
To show the predicted performance of the correlation vector machine algorithm optimized by the particle swarm optimization, an example is illustrated here. FIG. 3 is a schematic diagram of a Sallen-Key band-pass filter circuit, where V isoutThe initial value of R2 in a fault-free interval is 3 × (1 + 0.05) =3.15k omega, the resistance value of R2 in the fault is 3 ×.5=4.5 k omega, if the resistance value of R2 in the fault is 5 omega, the parameter value of R2 after 270 time points reaches the fault value of 4.5k omega from the initial value of 3.15k omega, and the invention sets R2 to rise from the initial value of 3.15k omega until the resistance value is 4.75 k omega for conveniently viewing the prediction effect, and the total time points are 320.
The failure prediction feature vector when R2 is at the initial value and the failure prediction feature vector at 320 time points are extracted, the health of the 320 time points is calculated and shown in fig. 4 in increments of time points, wherein the resistance value reaches 4.5k Ω at 270 th time point, R2 fails, and the health failure threshold is 0.999765181360359. The particle swarm optimization adopts the data of the first 100 time points as training data, kernel function parameter factors of a correlation vector machine algorithm are optimized and selected, the value obtained after optimization is 0.443725069454054, the optimized correlation vector machine algorithm is used for predicting the fault of the analog circuit, the effect comparison of the predicted data and the effect of the original data is shown in figure 5, the time when the fault is predicted is the 268 th time point, the error is-2 time points, and the Mse between the predicted data and the original data of each time point is 5.3340 e-11. Since the kernel function parameter factors obtained by optimization are close to 0.45, the occurrence time of the fault is predicted by using 0.1, 0.2, 0.3, 0.4, 0.45, 0.5, 0.6, 0.7, 0.8, 0.9 and 1 as the kernel function parameter factors respectively, and the prediction results are shown in table 1. Where >320 indicates that the maximum range of data has been exceeded and >50 represents that the maximum range of errors has been exceeded.
TABLE 1 results predicted at parameter factors of 0.1, 0.2, 0.3, 0.4, 0.45, 0.5, 0.6, 0.7, 0.8, 0.9 and 1
| Parameter factor | Predicting a point in time at which a fault occurs | Error of the measurement | Mse |
| 0.1 | 143 | -127 | 1.1876e-08 |
| 0.2 | 186 | -84 | 4.8176e-09 |
| 0.3 | 207 | -63 | 2.6312e-09 |
| 0.4 | 245 | -25 | 3.9886e-10 |
| 0.45 | 213 | -57 | 2.4730e-09 |
| 0.5 | 228 | -42 | 1.4181e-09 |
| 0.6 | 258 | -12 | 3.4975e-10 |
| 0.7 | 286 | 16 | 1.7554e-10 |
| 0.8 | 307 | 37 | 3.9992e-10 |
| 0.9 | >320 | >50 | 7.5851e-10 |
| 1 | >320 | >50 | 1.1708e-09 |
From the results in table 1, it can be known that the particle swarm optimization optimizes the correlation vector machine algorithm of the parameter factors, the time point of failure prediction is the most accurate, the prediction error is the minimum, only-2 time points are provided, and the Mse is the minimum. And when the parameter factors of the correlation vector machine algorithm are set to be 0.1, 0.2, 0.3, 0.4, 0.45, 0.5, 0.6, 0.7, 0.8, 0.9 and 1, the predicted error and the Mse are both larger than those of the correlation vector machine algorithm which optimizes the parameter factors through the particle swarm optimization and is provided by the invention. The result shows that the prediction performance of the correlation vector machine algorithm is obviously improved after the kernel function width factor is optimized by the particle swarm optimization.
The method has simple characteristic quantity extraction and small calculation amount, so the method can be used for a real-time system and a non-real-time system.
The method can also be applied to the capacitor and the inductor due to the characteristic that the parameter value changes in the aging process of the capacitor and the inductor.
Because the nonlinear analog circuit has the characteristic of parameter value change in the element aging process, and meanwhile, the nonlinear analog circuit can also extract the sweep frequency response signal of the test node, the method can predict the fault of the linear analog circuit and can also predict the fault of the nonlinear analog circuit.
Claims (8)
1. A method for predicting the failure of an analog circuit is characterized by comprising the following steps:
(1) carrying out Monte Carlo analysis on each element of the analog circuit in a fault-free interval, extracting a test node signal, carrying out wavelet packet transformation denoising processing on the extracted signal, and extracting signal energy of each frequency band, wherein the test node signal is a branch voltage;
(2) normalizing the extracted frequency band signal energy to obtain a fault diagnosis characteristic vector;
(3) taking the fault diagnosis feature vector as training data to train a BP neural network;
(4) extracting node signals of a circuit to be detected during working, carrying out wavelet packet transformation and normalization to generate corresponding fault diagnosis characteristic vectors, and judging the types of faults with trends by using a BP neural network;
(5) extracting a swept frequency response signal of a tested circuit test node of an initial value used for simulating circuit fault prediction by an element to form a fault prediction characteristic vector of the initial value used for simulating circuit fault prediction, wherein the response signal is branch voltage;
the definition of the initial value is: the initial value for analog circuit failure prediction is equal to the nominal value x (1 + upper normal tolerance) when the component fails in a forward direction away from the nominal value, and equal to the nominal value x (1 + lower normal tolerance) when the component fails in a reverse direction away from the nominal value;
(6) extracting a frequency sweep response signal of a test node of the analog circuit to be tested during working according to a fixed time interval so as to form a fault prediction characteristic vector of a time point sequence of an element;
(7) calculating the cosine angle distance between the element fault prediction characteristic vector extracted according to a fixed time interval and the fault prediction characteristic vector of an initial value used for simulating circuit fault prediction of the element, and representing the health degree of the element at different time points to generate a corresponding health degree sequence;
(8) calculating a health threshold when the element fails;
(9) based on the element health degree sequence and time point sequence data, optimizing and selecting a kernel function width factor of a correlation vector machine algorithm by applying a particle swarm algorithm;
(10) and (3) performing fault prediction on the analog circuit by using a correlation vector machine algorithm optimized by a particle swarm algorithm.
2. The method of claim 1, wherein in step (1), the no-fault interval is [50%,1+ lower normal tolerance ] of the nominal value of the analog circuit component and [1+ upper normal tolerance, 150% ] of the nominal value.
3. The analog circuit failure prediction method of claim 1, wherein in the step (4), the failure category with tendency refers to a failed component, and a direction in which the component deviates from an initial value for analog circuit failure prediction when a failure occurs.
4. The method according to claim 1, wherein in step (6), the fixed interval time points of the sweep response signal of the test node of the analog circuit under test are extracted sequentiallyT 1 , T 2 ,…, T n The time point sequence thus constituted is [ 2 ]T 1 , T 2 ,…, T n ]And is the total number of time points.
5. The analog circuit failure prediction method of claim 1, wherein in the step (7), the cosine angular distance is calculated by:
；
wherein,characterized by a fault prediction feature vector of initial values for analog circuit fault prediction,representing a vectorTo middlejThe characteristics of the device are as follows,，mis the number of features in the extracted set of feature vectors;characterized by the firstiThe component failure prediction feature vectors extracted at each time point,representing a vectorTo middlejThe characteristics of the device are as follows,，is the total number of extracted feature quantities, and is also the total number of time points;
degree of healthThe health degree sequence generated by the element failure prediction feature vector extracted at regular time intervals is [ alpha ]H 1 , H 2 ,…, H n ]The corresponding time point sequence is [ 2 ]T 1 , T 2 ,…, T n ]。
6. The analog circuit failure prediction method of claim 1, wherein in the step (8), the health threshold at the time of the element failure is calculated by: calculating the cosine angular distance of the final deviation value in the fault-free interval of the deviation direction of the element, namely the health degree threshold value of the element in the deviation direction with fault; the health threshold when a component fails may also be referred to as a health failure threshold for short.
7. The analog circuit fault prediction method according to claim 1, wherein in the step (9), the step of applying the particle swarm optimization to optimally select the kernel function width factor of the correlation vector machine algorithm comprises:
(1) initializing particle swarm algorithm parameters including position, speed, optimizing range and iteration times, wherein the width factor is mapped to the position of the particle;
(2) calculating the fitness of each particle, and obtaining the individual optimal position of each particle and the global optimal position of the particle swarm according to the fitness;
(3) updating the speed and the position of each particle;
(4) and (5) repeating the steps (2) and (3) until the iteration is finished, and outputting the result.
8. The analog circuit failure prediction method of claim 1, wherein, in the step (10),
the fault prediction of the analog circuit comprises two modes:
(1) predicting a failure occurrence time point of the element;
(2) and predicting whether the element fails at a certain future time point.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410088347.6A CN103824135B (en) | 2014-03-11 | A kind of analog circuit fault Forecasting Methodology |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410088347.6A CN103824135B (en) | 2014-03-11 | A kind of analog circuit fault Forecasting Methodology |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103824135A CN103824135A (en) | 2014-05-28 |
| CN103824135B true CN103824135B (en) | 2016-11-30 |
Family
ID=
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101221213A (en) * | 2008-01-25 | 2008-07-16 | 湖南大学 | Analogue circuit fault diagnosis neural network method based on particle swarm algorithm |
| CN102222151A (en) * | 2011-07-21 | 2011-10-19 | 电子科技大学 | Analog circuit fault prediction method based on ARMA (Autoregressive Moving Average) |
| CN102636740A (en) * | 2012-04-18 | 2012-08-15 | 南京航空航天大学 | Method for predicting faults of power electronic circuit based on FRM-RVM (fuzzy rough membership-relevant vector machine) |
| CN102662142A (en) * | 2012-03-15 | 2012-09-12 | 南京航空航天大学 | Prediction method for multi-parameter identification fault of power electronic circuit based on RVM-QNN |
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101221213A (en) * | 2008-01-25 | 2008-07-16 | 湖南大学 | Analogue circuit fault diagnosis neural network method based on particle swarm algorithm |
| CN102222151A (en) * | 2011-07-21 | 2011-10-19 | 电子科技大学 | Analog circuit fault prediction method based on ARMA (Autoregressive Moving Average) |
| CN102662142A (en) * | 2012-03-15 | 2012-09-12 | 南京航空航天大学 | Prediction method for multi-parameter identification fault of power electronic circuit based on RVM-QNN |
| CN102636740A (en) * | 2012-04-18 | 2012-08-15 | 南京航空航天大学 | Method for predicting faults of power electronic circuit based on FRM-RVM (fuzzy rough membership-relevant vector machine) |
Non-Patent Citations (2)
| Title |
|---|
| Particle swarm optimization based RVM classifier for non-linear circuit fault diagnosis;Cheng Gao et al.;《Journal of Central South University》;20120128;第19卷(第2期);第459-464 * |
| 基于小波与LS-SVM集成的模拟电路故障检测;彭四海;《电子设计工程》;20130531;第21卷(第10期);第119-122页 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101477172B (en) | Analogue circuit fault diagnosis method based on neural network | |
| CN107992665B (en) | Online fault diagnosis and analysis method for alternating current filter of extra-high voltage converter station | |
| CN103514366B (en) | Urban air quality concentration monitoring missing data recovering method | |
| KR101822829B1 (en) | Method of identifying partial discharge and noise of switchgear using machine learning | |
| WO2016107246A1 (en) | Wavelet noise reduction and relevance vector machine-based method for predicting remaining life of lithium battery | |
| CN109323754B (en) | Train wheel polygon fault diagnosis and detection method | |
| CN105572572B (en) | Analog-circuit fault diagnosis method based on WKNN-LSSVM | |
| CN112051480A (en) | Neural network power distribution network fault diagnosis method and system based on variational modal decomposition | |
| CN112685961A (en) | Method and system for predicting remaining service life of analog circuit | |
| CN116226646B (en) | Method, system, equipment and medium for predicting health state and residual life of bearing | |
| CN101739819A (en) | Method and device for predicting traffic flow | |
| CN103675525A (en) | DC-DC converter health monitoring and fault prediction method based on multiple SVDD models | |
| CN110618353A (en) | Direct current arc fault detection method based on wavelet transformation + CNN | |
| CN105005294A (en) | Real-time sensor fault diagnosis method based on uncertainty analysis | |
| Baumann et al. | Impulse test fault diagnosis on power transformers using Kohonen's self-organizing neural network | |
| CN105866664A (en) | Intelligent fault diagnosis method for analog circuit based on amplitude frequency features | |
| CN106771938A (en) | A kind of solid insulation ring main unit Partial Discharge Pattern Recognition Method and device | |
| CN106301610A (en) | The adaptive failure detection of a kind of superhet and diagnostic method and device | |
| CN106021671A (en) | Circuit health ranking evaluation method in combination with dependency relation and gray clustering technology | |
| CN111611654A (en) | Fatigue prediction method, device and equipment for riveted structure and storage medium | |
| Liao et al. | Recognizing noise-influenced power quality events with integrated feature extraction and neuro-fuzzy network | |
| Sridhar et al. | Detection and classification of power quality disturbances in the supply to induction motor using wavelet transform and neural networks | |
| CN103824135B (en) | A kind of analog circuit fault Forecasting Methodology | |
| CN113866614A (en) | Multi-scenario user side low-voltage direct-current switch arc fault diagnosis method and device | |
| Papadopoulos et al. | Online parameter identification and generic modeling derivation of a dynamic load model in distribution grids |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |