Summary of the invention
Object of the present invention is exactly for the deficiencies in the prior art, there is provided a kind of utilize relative amplitude-phase analysis, can the analog circuit fault detection method of high-quality discrimination circuit fault, accuracy of detection is high, and fault coverage is high, to catastrophic type and parameter type fault all effective.
For achieving the above object, technical scheme of the present invention is as follows:
Ultimate principle of the present invention is: the fault effect in mimic channel can characterize with faulty circuit signal and the relative energy of normal circuit signal and the change of time delay.Catastrophic type fault in mimic channel is usually expressed as the marked change of relative energy between faulty circuit signal and normal circuit signal, and parameter type fault is often insensitive to the relative energy change of faulty circuit signal and normal circuit signal, but to the relative time-delay sensitive between faulty circuit signal and normal circuit signal.The present invention is based on this true, then combine multiple this efficient theoretical tool of mutual wavelet transformation, high-quality fault detect can be brought.
A mimic channel, wavelet transformation is mutually carried out again to the sample sequence of its faulty circuit and normal circuit, so its relative amplitudes represents that faulty circuit exports the relative energy change exported normal circuit, and relative phase values represents that faulty circuit exports the relative phase offset exported normal circuit.Multiple-mutual wavelet transformation of relative amplitude analytical approach utilization can extract the Relative Signal Amplitude between faulty circuit signal and normal circuit signal on different time frequency domain yardsticks, and Relative Signal Amplitude shows as the relative energy change between faulty circuit signal and normal circuit signal.Utilize relative amplitude analytical approach effectively can detect fault to relative energy sensitive in mimic channel.On the other hand, multiple-mutual wavelet transformation of relative phase analytical approach utilization can extract the relative phase information between faulty circuit signal and normal circuit signal on different time frequency domain yardsticks, and relative phase information shows as the time delay between faulty circuit signal and normal circuit signal.Utilize relative phase analytical approach effectively can detect fault to time delay sensitive in mimic channel.The fault detection method that the present invention proposes requires that unknown failure sequence and non-fault sequence need to sample in the same triggering sampling time, then could be used for extracting the relative phase offset between sample sequence and energy variation.Present-day data collection device all has triggering and timing function, and can complete this requirement well, this is the support that the inventive method obtains that in engineering practice practicality provides objective material conditions.
Basic theories of the present invention is as follows:
The sample sequence of a circuit is X (n), and its continuous wavelet transform may be defined as the convolution of X (n) and morther wavelet ψ (t), namely
In above formula, * represents complex conjugate, and Δ t represents sampling interval, and s represents dimensions in frequency, and n is time shifting.Multiple continuous wavelet transform requires that ψ (t) is for multiple morther wavelet.
The multiple mutual wavelet transformation of sample sequence X and Y is defined as:
relative amplitude
can think sequence X and the Y relative energy at a certain dimensions in frequency s and time shifting n;
phase place
sequence X and the Y relative phase at a certain dimensions in frequency s and time shifting n can be thought.Here, the relative phase of burst means the time delay between signal.The relative energy that analog circuit fault shows as between fault sample sequence and circuit normal sample sequence changes and relative phase offset.Multiple mutual wavelet transformation of the present invention adopts Morlet morther wavelet, or Paul morther wavelet.
In analog circuit fault diagnosing, it is the sine wave of F1 that circuit-under-test input is generally a frequency, and circuit exports response and is usually confined within frequency response range F2.According to the relation of wavelet scale in wavelet analysis and equivalent Fourier frequency (see formula below 1. 2.), the wavelet analysis yardstick SF corresponding with frequency response F2 can be obtained.
①
②
Formula 1. 2. in s be the frequency domain yardstick of wavelet transformation, f is equivalent Fourier frequency, ω
0for the angular frequency of morther wavelet, general ω
0=6, m is the exponent number of Paul small echo, general m=3.
be expressed as relative amplitude on different frequency yardstick s and time shifting n of circuit normal sequence and failure sequence and phase value.Usually small echo dimensions in frequency SF is concentrated on because circuit exports response, so can be from
middle extraction sensitive information
In order to overcome the edge effect that wavelet transformation brings, only calculate cone of influence (COI) extraneous relative amplitude and phase value.Then sensitive information extraction formula can be expressed as:
The extracting method of sensitive information is provided below with an example forms.
Suppose a linear analogue circuit-under-test, its input stimulus is the sine wave of a 1KHz, and circuit exports response and stores under the sampling rate of 100KHz, each Waveform storage 2048 points.We adopt ω
0=6Morlet morther wavelet carries out again mutually wavelet transformation to output sample sequence, mother wavelet function ψ (t) as shown in the formula:
Because circuit is linear circuit, it is also 1KHz that its circuit exports response frequency scope, so according to Morlet wavelet scale and equivalent Fourier frequency relation, and the dimensions in frequency SF of wavelet transformation when can to extrapolate equivalent Fourier frequency be 1KHz:
Suppose V-Fault and V-Normal be circuit under certain failure condition and all elements be that circuit under its nominal value exports response, get 2048 sampled points.Multiple mutual small echo dimensions in frequency S, adopt following two formula according to document suggestion, the selection of S must cover SF.The value of S is [s
0, s
1..s
j].
So multiple mutual wavelet transformation
for the complex element matrix of J*2048, wherein dimensions in frequency is S, and length is J, and time shifting is [1,2 ..2048], and length is 2048, then
It is the complex element matrix of a 1*2048.
Because sample sequence is finite length, directly wavelet transformation is mutually carried out again to sample sequence, to cause, in the beginning of dimensions in frequency S and ending phase, larger error occurs.A common optimization solution, for carry out zero padding operation to sample sequence before carrying out wavelet transformation, is then removed after wavelet transform.Zero padding operation is carried out to sample sequence here, make sequence length reach original series length the most close 2
nindividual, because this reducing edge effect and accelerating Fourier transform process.Cone of influence is defined as again the mutually e times die-away time of wavelet transformation auto-correlation energy spectrum under each dimensions in frequency S.E times of die-away time guarantees to make edge effect decline e
-2doubly, ensure that edge effect is negligible.
for the 1*2048 sequence under dimensions in frequency SF, in order to overcome the impact of edge effect, according to the data of only getting cone of influence inside said above.Supposing beginning under SF and terminating displacement die-away time is N1, N2, then the sample sequence of extraction time section N1-N2:
So
be the sample sequence of 2048-N1-(2048-N2) length, this sequence is the fault characteristic information abstraction sequence useful to the relative amplitude-phase analysis method of the present invention's proposition.
Specifically, the fault detection method of a kind of mimic channel that the present invention proposes, concrete steps are as follows:
(1) each component parameter of tested mimic channel is set to nominal parameters, emulates this tested mimic channel, under obtaining nominal parameters, circuit exports response sequence V-Normal.Carry out in HSPICE the emulation of this tested mimic channel, circuit is input as sine wave.V-Normal sequence is used as the benchmark of wavelet transformation mutually again.
(2) in the range of tolerable variance of each components and parts nominal parameters of tested mimic channel, (circuit components parameter meets Gaussian distribution, within the scope of 1 times of its standard deviation, i.e. " 1Sigma "), Monte Carlo simulation is carried out to this tested mimic channel, obtains exporting response sequence with the circuit of Monte Carlo simulation number of times same number.Monte Carlo simulation number of times is M time, and it is M that the circuit obtained exports response sequence, and wherein V-Monte (i) is expressed as the circuit output response sequence of i-th emulation, 1≤i≤M.
(3) circuit under the nominal parameters obtained in step (1) is exported response sequence and carry out again wavelet transformation mutually, namely
And therefrom extract sensitive information, obtain relative amplitude reference value sequence
with relative phase reference value sequence
Export response sequence by what obtain in step (2) with the circuit of Monte Carlo simulation number of times same number, export response sequence with circuit under the nominal parameters that obtains in step (1) respectively and carry out again wavelet transformation mutually, namely
and therefrom extract sensitive information, obtain the relative amplitude simulation value sequence with aforementioned Monte Carlo simulation number of times same number
and with the relative phase simulation value sequence of aforementioned Monte Carlo simulation number of times same number
(4) by obtain in step (3) with the relative amplitude simulation value sequence A MP of aforementioned Monte Carlo simulation number of times same number
montei (), respectively with the relative amplitude reference value sequence A MP that obtains in step (3)
referencebe normalized, obtain exporting with the normal circuit of aforementioned Monte Carlo simulation number of times same number responding relative amplitudes AMP
rEF(i).
By obtain in step (3) with the relative phase simulation value sequence A RG of aforementioned Monte Carlo simulation number of times same number
montei (), respectively with the relative phase reference value sequence A RG that obtains in step (3)
referencebe normalized, obtain exporting with the normal circuit of aforementioned Monte Carlo simulation number of times same number responding relative phase values ARG
rEF(i).
Normalized formula is as follows:
Wherein
(5) exporting with the normal circuit of aforementioned Monte Carlo simulation number of times same number of obtaining in step (4) responds relative amplitudes AMP
rEFin (i), select numerical value the maximum as maximum relative amplitudes max (AMP
rEF), select numerical value reckling as minimum relative amplitudes min (AMP
rEF).
Exporting with the normal circuit of aforementioned Monte Carlo simulation number of times same number of obtaining in step (4) responds relative phase values ARG
rEFin (i), select numerical value the maximum as maximum relative phase values max (ARG
rEF), select numerical value reckling as minimum relative phase values min (ARG
rEF).
Above-mentioned steps (1)-(5) are the circuit simulation stage, utilize Monte-Carlo to emulate to obtain the scope of relative amplitude that normal circuit (within the scope of parameter tolerances) responds and phase place, determine minimax relative amplitudes and minimax relative phase values.
(6) the tested mimic channel of unknown failure is surveyed, obtain the actual measurement of unknown circuit and export response sequence V-Measured.
(7) circuit under the nominal parameters obtained in step (1) is exported the unknown circuit obtained in response sequence V-Normal and step (6) to survey and export response sequence V-Measured and carry out again mutual wavelet transformation, namely
And therefrom extract sensitive information, obtain the actual measurement relative amplitude value sequence of unknown failure circuit-under-test
with the actual measurement relative phase value sequence of unknown failure circuit-under-test
(8) the relative amplitude reference value sequence A MP will obtained in step (3)
referencewith the actual measurement relative amplitudes sequence A MP of the unknown failure circuit-under-test obtained in step (7)
measuredbe normalized, obtain unknown circuit and export response relative amplitudes AMP
unknown.
By the relative phase reference value sequence A RG obtained in step (3)
referencewith the actual measurement relative phase values sequence A RG of the unknown failure circuit-under-test obtained in step (7)
measuredbe normalized, obtain unknown circuit and export response relative phase values ARG
unknown.
Normalized formula is as follows:
(9) the unknown circuit obtained in step (8) is exported response relative amplitudes AMP
unknown, with the maximum relative amplitudes max (AMP obtained in step (5)
rEF) and minimum relative amplitudes min (AMP
rEF) compare; The unknown circuit obtained in step (8) is exported response relative phase values ARG simultaneously
unknown, with the maximum relative phase values max (ARG obtained in step (5)
rEF) and minimum relative phase values min (ARG
rEF) compare.
If unknown circuit exports response, relative amplitudes is more than or equal to minimum relative amplitudes, is less than or equal to maximum relative amplitudes simultaneously, and unknown circuit output response relative phase values is more than or equal to minimum relative phase values, is less than or equal to maximum relative phase values (i.e. min (AMP simultaneously
rEF)≤AMP
unknown≤ max (AMP
rEF), min (ARG simultaneously
rEF)≤ARG
unknown≤ max (ARG
rEF)), the so tested mimic channel of this unknown failure then non-fault.
If unknown circuit exports response, relative amplitudes is less than minimum relative amplitudes, or is greater than maximum relative amplitudes (i.e. AMP
unknown< min (AMP
rEF) or AMP
unknown> max (AMP
rEF)), so then there is fault (relative amplitude fault) in the tested mimic channel of this unknown failure; If unknown circuit exports response, relative phase values is less than minimum relative phase values, or is greater than maximum relative phase values (i.e. ARG
unknown< min (ARG
rEF) or ARG
unknown> max (ARG
rEF)), so then there is fault (relative phase fault) in the tested mimic channel of this unknown failure.
Compared with prior art, the invention has the beneficial effects as follows: utilize relative amplitude-phase analysis to detect analog circuit fault, not singly close from the amplitude versus frequency of circuit-under-test test response to fasten differentiation fault, also in conjunction with the phase-frequency characteristic of circuit-under-test test response, accuracy of detection is high, fault coverage is high, to catastrophic type and parameter type fault all effective.
Embodiment
Below in conjunction with accompanying drawing, the preferred embodiments of the present invention are further described.
Embodiment 1
As shown in Figure 1, Figure 2, Figure 3 shows.Choose international benchmark leapfrog Circuit verification analog circuit fault detection method of the present invention.The fault detection method step of described mimic channel is as follows:
(1) each component parameter of tested mimic channel is set to nominal parameters, the R namely in Fig. 2
1~ R
13=10K Ω, C
1=C
4=10nF, C
2=C
3c in=20nF, Fig. 3
1=6pF.Emulate in HSPICE this tested mimic channel, the 1kHz of input VIN to be amplitude be 3V is sinusoidal wave, exports VOUT and samples under 100K sps, stores 2048 points at every turn, and under obtaining nominal parameters, circuit exports response sequence.
(2) in the range of tolerable variance of each components and parts nominal parameters of tested mimic channel, 2000 Monte Carlo simulations are carried out to this tested mimic channel, obtain 2000 circuit and export response sequence.
(3) circuit under the nominal parameters obtained in step (1) is exported response sequence and carry out again wavelet transformation mutually, and therefrom extract sensitive information, obtain relative amplitude reference value sequence and relative phase reference value sequence.
2000 circuit obtained in step (2) are exported response sequence, export response sequence with circuit under the nominal parameters that obtains in step (1) respectively and carry out again wavelet transformation mutually, and therefrom extract sensitive information, obtain 2000 relative amplitude simulation value sequences and 2000 relative phase simulation value sequences.
(4) 2000 the relative amplitude simulation value sequences will obtained in step (3), are normalized with the relative amplitude reference value sequence that obtains in step (3) respectively, obtain 2000 normal circuit and export response relative amplitudes.
By 2000 the relative phase simulation value sequences obtained in step (3), be normalized with the relative phase reference value sequence that obtains in step (3) respectively, obtain 2000 normal circuit and export response relative phase values.
(5) 2000 normal circuit obtained in step (4) export in response relative amplitudes, select numerical value the maximum as maximum relative amplitudes, select numerical value reckling as minimum relative amplitudes.
2000 normal circuit obtained in step (4) export in response relative phase values, select numerical value the maximum as maximum relative phase values, select numerical value reckling as minimum relative phase values.
Carry out above-mentioned steps (1)-(5) with Morlet morther wavelet and Paul morther wavelet respectively, then obtain the relative amplitude of normal circuit response and the scope of relative phase, as shown in table 1.Can find out, in time adopting Morlet morther wavelet, the relative amplitude scope absolute value of normal circuit is about 25.8, and adopt Paul morther wavelet relative amplitude scope to be about 9.0, thus can draw and adopt Morlet morther wavelet in relative amplitude analysis, to have better fault resolution than Paul small echo.And for relative phase analysis, its normal response scope absolute value is all about 18.2, the effect that these two kinds of morther wavelet are analyzed relative phase is suitable.
|
Minimum value |
Maximal value |
Relative amplitudes (Morlet morther wavelet) |
-13.498259 |
12.343993 |
Relative phase values (Morlet morther wavelet) |
-8.895469 |
9.295972 |
Relative amplitudes (PAUL morther wavelet) |
-4.740685 |
4.336003 |
Relative phase values (PAUL morther wavelet) |
-8.885899 |
9.293368 |
Table 1 leapfrog circuit normal response relative amplitude and relative phase bounds
(6) 10 known busts and 14 known parameter type faults are injected leapfrog circuit respectively, obtain 24 actual measurements accordingly and export response sequences.
(7) actual measurement corresponding with 10 busts obtained in step (6) is exported response sequence to export response sequence Morlet morther wavelet with circuit under the nominal parameters that obtains in step (1) respectively and carry out again wavelet transformation mutually, and therefrom extract sensitive information, obtain 10 actual measurement relative amplitude value sequences and 10 actual measurement relative phase value sequences.
The actual measurement corresponding with 14 parameter type faults obtained in step (6) is exported response sequence to export response sequence with circuit under the nominal parameters that obtains in step (1) respectively and carry out again wavelet transformation mutually, multiple mutual wavelet transformation carries out with Morlet morther wavelet and Paul morther wavelet respectively, and therefrom extract sensitive information, obtain 28 actual measurement relative amplitude value sequences and 28 actual measurement relative phase value sequences.
(8) 10 that obtain in step (7) actual measurement relative amplitude value sequences are normalized with the relative amplitude reference value sequence that obtains in step (3) respectively, obtain the output response relative amplitudes of 10 known busts, as shown in table 2.10 that obtain in step (7) actual measurement relative phase value sequences are normalized with the relative phase reference value sequence that obtains in step (3) respectively, obtain the output response relative phase values of 10 known busts, as shown in table 2.
Table 2 leapfrog circuit typical disaster fault
28 that obtain in step (7) actual measurement relative amplitude value sequences are normalized with the relative amplitude reference value sequence that obtains in step (3) respectively, obtain the output response relative amplitudes of 28 known parameters type faults, as shown in table 3.28 that obtain in step (7) actual measurement relative phase value sequences are normalized with the relative phase reference value sequence that obtains in step (3) respectively, obtain the output response relative phase values of 28 known parameters type faults, as shown in table 3.
Table 3 leapfrog circuit canonical parameter type fault
(9) the output response relative amplitudes will obtained in step (8), compares with the maximum relative amplitudes obtained in step (5) and minimum relative amplitudes; The output response relative phase values simultaneously will obtained in step (8), compares with the maximum relative phase values obtained in step (5) and minimum relative phase values.
In table 2, C4-Open is that breaking phenomena appears in element C4, and C4-Short is that short circuit phenomenon appears in element C4.AMP (Morlet) represents the relative amplitudes of corresponding fault under Morlet morther wavelet, ARG (Morlet) represents the relative phase values of corresponding fault under Morlet morther wavelet, and OBS (Morlet) can by this fault detect out under representing employing relative amplitude or relation analysis.
For C4-Open fault, the analysis of employing relative phase can well by this fault detect out, and it shows as faulty circuit sample sequence and normal circuit sample sequence has an obvious time delay.For R10-Open fault, it analyzes responsive to relative amplitude, and analyzes insensitive to relative phase, and it shows as faulty circuit sample sequence has significant change relative to normal circuit sample sequence energy.For faults such as C4-Short, R10-Short and R12-Short, can detect with relative amplitude and relative phase analysis, its fault not only shows the change of energy between sample sequence, has an obvious relative time-delay simultaneously.It is relatively large that bust often shows that circuit exports response energy variation, and such as R12-Short, R11-Open fault, it is almost nil that its sample sequence shows as sampled output signal amplitude in time domain.And for C4-Open, C1-Open fault, its sample sequence amplitude in time domain almost exports the same with normal circuit, but it significantly shows that an obvious time delay appears in faulty circuit and normal circuit outlet chamber.
In table 3, C1-6Sigma is element C1 minimizing 6sigma, C2+6Sigma is that element C2 increases 6sigma.AMP (Morlet) represents the relative amplitudes of corresponding fault under Morlet morther wavelet, ARG (Morlet) represents the relative phase values of corresponding fault under Morlet morther wavelet, AMP (Paul) represents the relative amplitudes of corresponding fault under Paul morther wavelet, and ARG (Paul) represents the relative phase values of corresponding fault under Paul morther wavelet.
When C1 reduces 6sigma, relative amplitudes overrun upper boundary values 12.343993, therefore C1-6Sigma fault can detect with relative amplitude analysis.When C3 increases 6sigma, relative phase values overrun lower border value-8.895469, therefore C3+6Sigma fault can detect with relative phase analysis.C2+6Sigma and C3-6Sigma fault can be detected by Morlet morther wavelet and Paul morther wavelet simultaneously, and the selection major effect of morther wavelet shows fault resolution aspect.For C1-6Sigma, R1-6Sigma and R2-6Sigma fault, it is more more responsive than relative phase analysis to relative amplitude analysis.Meanwhile, for C3+6Sigma, C4+6sigma and R6-6sigma fault, it analyzes insensitive to relative amplitude, and can detect this type of fault with relative phase analysis.For C2+6Sigma and R7-6sigma fault, detected by it can be analyzed by relative amplitude and relative phase, wherein C2+6Sigma fault is more responsive to relative phase analysis, and R7-6sigma fault is more responsive to relative amplitude analysis.