Redundancy signal deviation fault diagnosis grading threshold value calculation method based on difference value joint distribution and channel switching control process of redundancy signal deviation fault diagnosis grading threshold value calculation method

Abstract

The invention discloses a redundancy signal deviation fault diagnosis grading threshold calculation method based on difference joint distribution and a channel switching control process thereof, and solves the problem of redundancy signal deviation identification by constructing a new fault identification method. And performing modeling calculation on the redundant system according to the statistical characteristic parameters of the signals. Based on a joint normal distribution theory of redundant signal difference values, according to actual operation requirements of equipment, a corresponding false alarm rate and a corresponding omission ratio are respectively calculated by difference value conditional probabilities, an optimal deviation fault threshold value is determined, and a corresponding program algorithm is compiled. Accordingly, a novel and optimized fault detection switching process is constructed, the functions of fault identification and signal channel switching are achieved, and the stability and reliability of operation of the water-turbine generator set are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal diagnosis, and in particular relates to a method for calculating a threshold value for fault diagnosis of a redundant signal deviation based on a joint distribution of differences and a channel switching control flow thereof. Background Art

[0002] In the control system of a hydro-turbine generator, the use of redundant voting for important signals can effectively improve the system reliability and fault tolerance performance. Since redundant input has the advantage of multiple backup signals, when a signal fails, the backup signal can be enabled to maintain the normal operation of the equipment and improve the reliability of the equipment. Therefore, fault detection of redundant signals is extremely important in the safe and stable operation of the equipment.

[0003] The existing signal deviation fault diagnosis methods are: 1. The same signal measurement value exceeds the minimum or maximum limit judgment method, which is suitable for identifying more serious signal disconnection or signal over-range faults. The fault identification range is limited, and only extreme situations can be identified. It cannot identify general signal deviation faults and is not universally applicable. This method does not give full play to the advantage that multiple signals can be compared with each other; 2. The difference judgment of multiple signals is adopted, but the setting of the difference fault diagnosis threshold mainly relies on experience and lacks an effective theoretical basis. Therefore, it is difficult to guarantee the effectiveness of the threshold setting. There is a hidden danger of failure of fault judgment, which affects the safe operation of the equipment. If the threshold is set too large, there is a risk of refusal to operate, that is, when a signal fault occurs, it cannot be detected and countermeasures are taken in time. The missed detection rate is too high, resulting in the situation that the signal fault has not been discovered for a long time, causing the equipment to run with the fault for a long time without being noticed, which seriously affects the safety of the equipment. If the threshold is set too small, there is a risk of false operation, that is, the signal is judged as a fault within the normal error range, that is, it is impossible to avoid the normal error fluctuation of the signal and alarm the false operation of the equipment. The false alarm rate is too high and the equipment frequently alarms.

[0004] Problems with existing technologies include:

[0005] 1. The method of judging whether the same signal measurement value exceeds the minimum or maximum limit is used. It is suitable for identifying more serious signal disconnection or signal over-range faults. The fault identification range is limited, and it cannot identify signal deviation faults. It is not universally applicable. 2. The multi-channel signal difference judgment method is used. The setting of the signal deviation fault diagnosis threshold mainly relies on experience, and there is a lack of effective theoretical calculation basis for false alarm rate and detection rate. It is difficult to avoid the situation where the fault judgment fails due to unreasonable threshold setting, affecting the safe operation of the equipment. Therefore, it is necessary to design a redundant signal deviation fault diagnosis hierarchical threshold calculation method based on the joint distribution of differences and its channel switching control process to solve the above problems. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a redundant signal deviation fault diagnosis hierarchical threshold calculation method based on the joint distribution of differences and its channel switching control process. The method aims to determine an optimal deviation fault threshold calculation method under the premise of taking into account refusal to operate and malfunction, and to construct a new fault identification method to solve the problem of redundant signal deviation identification. According to the statistical characteristic parameters of the signal, the redundant system is modeled and calculated; based on the joint normal distribution theory of redundant signal differences, according to the actual operation requirements of the equipment, the corresponding false alarm rate and missed detection rate are calculated respectively by the conditional probability of the difference, the optimal deviation fault threshold is determined, and the corresponding program algorithm is written; accordingly, a novel and optimized fault detection switching process is constructed to realize fault identification and signal channel switching functions, and improve the stability and reliability of the operation of the hydro-turbine generator set.

[0007] In order to achieve the above technical effects, the technical solution adopted by the present invention is:

[0008] The method for calculating the threshold value of redundancy signal deviation fault diagnosis based on the joint distribution of differences and the channel switching control process thereof include the following steps:

[0009] S1, establish the probability distribution of the sensor signal after filtering;

[0010] S2, construct the probability distribution of signal differences;

[0011] S3, derive the expression of fault recognition rate;

[0012] S4, derive the expression of fault misjudgment rate;

[0013] S5, calculate the confidence interval of the fault recognition rate;

[0014] S6, calculate the confidence interval of the fault misjudgment rate;

[0015] S7, fault judgment domain setting value calculation:

[0016] The fault misjudgment rate is set to once in N years, and the maximum allowable delay for fault identification is T; the execution cycle of the controller is τ milliseconds, and the number of data collection times S for N years of continuous operation is:

[0017]

[0018] Look up the normal distribution table and select the value of lower bound b>0,β=bσ, so that S(1-φ(β))×2=S(1-φ(bσ))×2<2;

[0019] Set the segment delay strategy:

[0020] According to the difference a, the error rate λ is given s, calculate the required delay period m, set different delay times respectively; when a>b, construct the solution equation of m, solve the delay of m periods through the program algorithm, and draw a table;

[0021] S8, constructing a signal fault switching program according to the decision domain setting value obtained in step 7, and completing the equipment fault diagnosis and switching control mode.

[0022] Preferably, in step S1, the specific method for establishing the probability distribution of the sensor signal after filtering is:

[0023] Let 1# sensor signal variable ξ1, 2# sensor signal variable ξ2, ..., k# sensor signal variable ξ k ; Random variables ξ1, ξ2, ..., ξ k Independent of each other and follow Gaussian distribution;

[0024] Given ξ1~N(μ1,σ1 2 ),ξ2~N(μ2,σ2 2 ), ..., After taking the mean filter, ξ 1i ,ξ 2i ,…,ξ ki Obey N(μ i ,σ i 2 ),i=1,…,n;

[0025] in, represents the mean filter signal of sensor j#; j = 1, 2, ..., k, ξ ji represents the i-th sampling value of the j# sensor signal; j = 1, 2, ..., k, i = 1, ..., n; n represents the number of sampling times of signal filtering processing;

[0026] Since the linear combination of a finite number of independent normal random variables still obeys the normal distribution, the random variable Obey Gaussian distribution;

[0027] From ξ1~N(μ1,σ1 2 ), so the calculated value of the characteristic function of the random variable ξ1 is:

[0028]

[0029] in, is the distribution density function of ξ1; let but Therefore, the calculated characteristic function value of Y is:

[0030]

[0031] Therefore The calculated characteristic function value of is:

[0032]

[0033] Available Similarly, That is, the sensor mean filter signal obeys the distribution:

[0034]

[0035] make j=1,2,…,k, then in, represents the mathematical expectation of the mean filter signal of k# sensors; represents the standard deviation of the mean filter signal of the k# sensor; μ j Represents the mathematical expectation of the j# sensor filtered signal.

[0036] Preferably, in step S2, the specific method for constructing the probability distribution of the signal difference is:

[0037] The difference between the k# sensor signal and the 1# sensor signal after averaging and filtering is The difference between the k# sensor signal and the k# sensor signal after averaging and filtering is , ..., the difference between the k# sensor signal and the (k-1)# sensor signal after mean filtering is Right now The probability distribution is

[0038]

[0039] make in:

[0040]

[0041] In the formula, is the column vector of the difference between the mean filtered signal of the k# sensor and the mean filtered signals of the 1# sensor, 2# sensor, ..., (k-1)# sensor respectively; is the mean filter signal of k# sensor; The k-order matrix A is:

[0042]

[0043] Substitution Calculate:

[0044]

[0045] The above matrix is ​​used to calculate:

[0046] Follows k-ary normal distribution in:

[0047]

[0048]

[0049]

[0050]

[0051]

[0052]

[0053]

[0054] From the above formula we can get:

[0055]

[0056] Simplified to:

[0057]

[0058] From the above formula, we can get:

[0059]

[0060] The above formula can be constructed:

[0061] The above formula can be converted to:

[0062]

[0063] Preferably, in step S3, the specific method for deriving the fault recognition rate expression is:

[0064] The fault recognition rate indicates the probability that a certain signal to be detected can be effectively detected through the difference information between this signal and other signals when a deviation fault occurs. That is, when the deviation of this signal exceeds the threshold, the conditional probability that the difference between this signal and other signals exceeds the threshold is calculated as follows:

[0065] Known Follows ternary normal distribution in is its subvector;

[0066] The mathematical expectation is The covariance is

[0067] The following is given Under the conditions The distribution density function is linearly transformed, and Then we have:

[0068]

[0069] Pick have to The required linear transformation is:

[0070]

[0071] For and Distribution density function, calculate:

[0072]

[0073]

[0074]

[0075] have

[0076] in,

[0077] Depend on have to:

[0078]

[0079] Depend on have to:

[0080]

[0081] make x2=t k ;

[0082] The function is:

[0083]

[0084] Transformed into:

[0085]

[0086] Preferably, in step S4, the specific method for deriving the fault misjudgment rate expression is:

[0087] The fault misjudgment rate refers to the probability that the signal to be detected is normal when the system determines that the signal to be detected has a deviation fault, but in fact the signal is normal, that is, when the difference between this signal and the rest of the signals exceeds the threshold, the signal is normal, that is, the first type of error; the specific calculation process is:

[0088] Known Follows k-ary normal distribution in is its subvector; the mathematical expectation is The covariance is Calculate at a given Under the conditions The distribution density function of is as follows:

[0089] Do a linear transformation, let but Pick Then there is The required linear transformation is:

[0090]

[0091] Transform the above formula into:

[0092]

[0093]

[0094]

[0095]

[0096] Further find out:

[0097]

[0098]

[0099] in Further find out: Right now:

[0100]

[0101] in

[0102]

[0103]

[0104]

[0105]

[0106]

[0107]

[0108] Based on the fact that the variance of the same model of each sensor is the same, that is, It can be concluded that:

[0109]

[0110] Further find out:

[0111]

[0112] make x2=x2; that is

[0113]

[0114] Further find out:

[0115]

[0116] in make:

[0117]

[0118] Further find out:

[0119]

[0120] Preferably, in step S5, the specific method for calculating the confidence interval of the fault recognition rate is: From formula (5), it can be obtained that:

[0121] ;make but

[0122]

[0123] When t k =β≥0, we have:

[0124]

[0125] When t k =-β<0, we have:

[0126]

[0127] ; Combining the above formula to calculate:

[0128]

[0129] Preferably, in step S6, the specific method for calculating the confidence interval of the fault misjudgment rate is:

[0130] make From formula (6), we can get:

[0131]

[0132] Further we can get:

[0133]

[0134] Segment calculation:

[0135] (1) t1=t2=…=t k-1 =α;

[0136]

[0137] ;make Then we have:

[0138]

[0139] (2) t1 = t2 = -α;

[0140]

[0141] make Then we have:

[0142]

[0143] Further we can get:

[0144]

[0145] Preferably, in step S7, the specific method for calculating the fault decision domain setting value is:

[0146] The fault misjudgment rate is set to once in N years, and the maximum allowable delay for fault identification is T; the execution cycle of the controller is τ milliseconds, and the number of data collection times S for N years of continuous operation is:

[0147]

[0148] Look up the normal distribution table and select the value of lower bound b>0, β=bσ, so that S(1-φ(β))×2=S(1-φ(bσ))×2<2; that is, infβs.tS(1-φ(βσ))×2<2;

[0149] Set the segment delay strategy:

[0150] According to the difference a, the error rate λ is given s, calculate the required delay period m, and set different delay times respectively:

[0151] Depend on

[0152] have to and (φ(ba)) 2 =γ;

[0153] Given the misjudgment rate λ s , calculate the required delay period m, then:

[0154]

[0155]

[0156]

[0157] Therefore

[0158]

[0159] The solution is:

[0160]

[0161] To obtain the minimum m; simplify it to:

[0162]

[0163] Further simplified to:

[0164]

[0165] φ -1 (λ s )<0,a>b or φ -1 (λ s )>0,(φ -1 (λ s )) 2 -4ab+4a 2 >0

[0166] When a>b, there is a real number solution, and the solution that conforms to engineering practice is:

[0167] φ -1 (λ s )<0,a>b

[0168] ; Use the program algorithm to solve the delay of m cycles and draw it into a table.

[0169] Preferably, in step S8, the fault judgment switching process is as follows:

[0170] S801, the current preferred output signal is the kth path, and the first determination process and the second determination process are respectively executed;

[0171] The first judgment process is:

[0172] 11. Subtract the i-th signal from the k-th signal mean to obtain the difference, and compare the difference with βσ, i=1,2,3,…,k-1; if all the differences are greater than βσ, proceed to the next step; otherwise, return to step S801;

[0173] 12. Introduce the k-th signal deviation fault alarm and exit the alternative signal pool;

[0174] 13. Select the current preferred output signal as the k-1th channel;

[0175] 14. Subtract the i-th signal from the mean of the k-1-th signal to obtain the difference, and compare the difference with βσ, i=1,2,3,…,k-2; if all the differences are greater than βσ, proceed to the next step; otherwise, return to step 13;

[0176] 15. Let k = k-1 and return to step 12;

[0177] 16. Repeat steps 12 to 15, and complete the last cycle when k=3;

[0178] 17, output the one with the larger value between the first and second paths as the preferred output signal;

[0179] The second determination process is:

[0180] 21. Subtract the i-th signal from the mean value of the k-th signal to obtain the difference, and compare the difference with βσ, i=1,2,3,…,k-1; if all the differences are less than -βσ, proceed to the next step; otherwise, return to step S801;

[0181] 22. Introduce the k-th signal deviation fault alarm and exit the alternative signal pool;

[0182] 23. Select the current preferred output signal as the k-1th channel;

[0183] 24. Subtract the i-th signal from the mean of the k-1-th signal to obtain the difference, and compare the difference with βσ, i=1,2,3,…,k-2; if all the differences are less than -βσ, proceed to the next step; otherwise, return to step 23;

[0184] 25, let k = k-1, and return to step 22;

[0185] 26. Repeat steps 22 to 25, and complete the last cycle when k=3;

[0186] 27, output the one with the larger value among the first and second paths as the preferred output signal.

[0187] The beneficial effects of the present invention are as follows:

[0188] The present invention can realize fault detection of sensor signal deviation, and design corresponding voting algorithm and fault switching process, so as to solve the problems inherent in traditional fault identification methods, ensure system reliability and effectiveness of output signals, and optimize the working performance of hydropower units. Compared with the traditional delay method, the delay time is fixed, which is irrational and does not meet the requirements of practical applications in both theoretical demonstration and practical verification. The present invention is based on the difference conditional distribution theory, and sets different precise corresponding delay times according to the different levels of signal difference, thereby improving the accuracy and sensitivity of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0189] Figure 1 It is a schematic diagram of the process in the present invention;

[0190] Figure 2 A schematic diagram of the decision domain setting process in the present invention;

[0191] Figure 3 It is a schematic diagram of the fault judgment switching process in the present invention. DETAILED DESCRIPTION

[0192] Embodiment 1:

[0193] like Figure 1 As shown, the calculation method of the threshold value for fault diagnosis of redundancy signal deviation based on the joint distribution of differences and the channel switching control flow thereof include the following steps:

[0194] S1, establish the probability distribution of the sensor signal after filtering;

[0195] S2, construct the probability distribution of signal differences;

[0196] S3, derive the expression of fault recognition rate;

[0197] S4, derive the expression of fault misjudgment rate;

[0198] S5, calculate the confidence interval of the fault recognition rate;

[0199] S6, calculate the confidence interval of the fault misjudgment rate;

[0200] S7, fault judgment domain setting value calculation:

[0201] The fault misjudgment rate is set to once in N years, and the maximum allowable delay for fault identification is T; the execution cycle of the controller is τ milliseconds, and the number of data collection times S for N years of continuous operation is:

[0202]

[0203] Look up the normal distribution table and select the value of lower bound b>0,β=bσ, so that S(1-φ(β))×2=S(1-φ(bσ))×2<2;

[0204] Set the segment delay strategy:

[0205] According to the difference a, the error rate λ is given s , calculate the required delay period m, set different delay times respectively; when a>b, construct the solution equation of m, solve the delay of m periods through the program algorithm, and draw a table;

[0206] S8, constructing a signal fault switching program according to the decision domain setting value obtained in step 7, and completing the equipment fault diagnosis and switching control mode.

[0207] Preferably, in step S1, the specific method for establishing the probability distribution of the sensor signal after filtering is:

[0208] Let 1# sensor signal variable ξ1, 2# sensor signal variable ξ2, ..., k# sensor signal variable ξ k ; Random variables ξ1, ξ2, ..., ξ k Independent of each other and follow Gaussian distribution;

[0209] Given ξ1~N(μ1,σ1 2 ),ξ2~N(μ2,σ2 2 ), ..., After taking the mean filter, ξ 1i ,ξ 2i ,…,ξ ki Obey N(μ i ,σ i 2 ),i=1,…,n;

[0210] in, represents the mean filter signal of sensor j#; j = 1, 2, ..., k, ξ ji represents the i-th sampling value of the j# sensor signal; j = 1, 2, ..., k, i = 1, ..., n; n represents the number of sampling times of signal filtering processing;

[0211] Since the linear combination of a finite number of independent normal random variables still obeys the normal distribution, the random variable Obey Gaussian distribution;

[0212] From ξ1~N(μ1,σ1 2 ), so the calculated value of the characteristic function of the random variable ξ1 is:

[0213]

[0214] in, is the distribution density function of ξ1; let but Therefore, the calculated characteristic function value of Y is:

[0215]

[0216] Therefore The calculated characteristic function value of is:

[0217]

[0218] Available Similarly, That is, the sensor mean filter signal obeys the distribution:

[0219]

[0220] make but in, represents the mathematical expectation of the mean filter signal of k# sensors; represents the standard deviation of the mean filter signal of the k# sensor; μ j Represents the mathematical expectation of the j# sensor filtered signal.

[0221] Preferably, in step S2, the specific method for constructing the probability distribution of the signal difference is:

[0222] The difference between the k# sensor signal and the 1# sensor signal after averaging and filtering is The difference between the k# sensor signal and the k# sensor signal after averaging and filtering is , ..., the difference between the k# sensor signal and the (k-1)# sensor signal after mean filtering is Right now The probability distribution is

[0223]

[0224] make in:

[0225]

[0226] In the formula, is the column vector of the difference between the mean filtered signal of the k# sensor and the mean filtered signals of the 1# sensor, 2# sensor, ..., (k-1)# sensor respectively; is the mean filter signal of k# sensor; The k-order matrix A is:

[0227]

[0228] Substitution Calculate:

[0229]

[0230] The above matrix is ​​used to calculate:

[0231] Follows k-ary normal distribution in:

[0232]

[0233]

[0234]

[0235]

[0236]

[0237]

[0238]

[0239] From the above formula we can get:

[0240]

[0241] Simplified to:

[0242]

[0243] From the above formula, we can get:

[0244]

[0245] The above formula can be constructed:

[0246]

[0247] The above formula can be converted to:

[0248]

[0249] Preferably, in step S3, the specific method for deriving the fault recognition rate expression is:

[0250] The fault recognition rate indicates the probability that a certain signal to be detected can be effectively detected through the difference information between this signal and other signals when a deviation fault occurs. That is, when the deviation of this signal exceeds the threshold, the conditional probability that the difference between this signal and other signals exceeds the threshold is calculated as follows:

[0251] Known Follows ternary normal distribution in is its subvector;

[0252] The mathematical expectation is The covariance is

[0253] The following is given Under the conditions The distribution density function is linearly transformed, and Then we have:

[0254]

[0255] Pick have to The required linear transformation is:

[0256]

[0257] For and Distribution density function, calculate:

[0258]

[0259]

[0260]

[0261] have

[0262] in,

[0263] Depend on have to:

[0264]

[0265] Depend on have to:

[0266]

[0267] make x2=t k ;

[0268] The function is:

[0269]

[0270] Transformed into:

[0271]

[0272] Preferably, in step S4, the specific method for deriving the fault misjudgment rate expression is:

[0273] The fault misjudgment rate refers to the probability that the signal to be detected is normal when the system determines that the signal to be detected has a deviation fault, but in fact the signal is normal, that is, when the difference between this signal and the rest of the signals exceeds the threshold, the signal is normal, that is, the first type of error; the specific calculation process is:

[0274] Known Follows k-ary normal distribution in is its subvector; the mathematical expectation is The covariance is Calculate at a given The distribution density function of ν2 under the condition of , the specific method is:

[0275] Do a linear transformation, let but Pick Then there is The required linear transformation is:

[0276]

[0277] Transform the above formula into:

[0278]

[0279]

[0280]

[0281]

[0282] Further find out:

[0283]

[0284]

[0285] in

[0286] Further find out: Right now:

[0287]

[0288] in

[0289]

[0290]

[0291]

[0292]

[0293]

[0294]

[0295] Based on the fact that the variance of the same model of each sensor is the same, that is, It can be concluded that:

[0296]

[0297] Further find out:

[0298]

[0299] make x2=x2; that is

[0300]

[0301] Further find out:

[0302]

[0303] in make:

[0304]

[0305] Further find out:

[0306]

[0307] Preferably, in step S5, the specific method for calculating the confidence interval of the fault recognition rate is: From formula (5), it can be obtained that:

[0308] ;make but

[0309]

[0310] When t k =β≥0, we have:

[0311]

[0312] When t k =-β<0, we have:

[0313]

[0314] ; Combining the above formula to calculate:

[0315]

[0316] Preferably, in step S6, the specific method for calculating the confidence interval of the fault misjudgment rate is:

[0317] make From formula (6), we can get:

[0318]

[0319] Further we can get:

[0320]

[0321] Segment calculation:

[0322] (1) t1 = t2 = ... = t k-1 =α;

[0323]

[0324] ;make Then we have:

[0325]

[0326] (2) t1 = t2 = -α;

[0327]

[0328] make Then we have:

[0329]

[0330] Further we can get:

[0331]

[0332] Preferably, in step S7, the specific method for calculating the fault decision domain setting value is:

[0333] The fault misjudgment rate is set to once in N years, and the maximum allowable delay for fault identification is T; the execution cycle of the controller is τ milliseconds, and the number of data collection times S for N years of continuous operation is:

[0334]

[0335] Look up the normal distribution table and select the value of lower bound b>0, β=bσ, so that S(1-φ(β))×2=S(1-φ(bσ))×2<2; that is, infβs.tS(1-φ(βσ))×2<2;

[0336] Set the segment delay strategy:

[0337] According to the difference a, the error rate λ is given s , calculate the required delay period m, and set different delay times respectively:

[0338] Depend on

[0339] have to and (φ(ba)) 2 =γ;

[0340] Given the misjudgment rate λ s , calculate the required delay period m, then:

[0341]

[0342]

[0343]

[0344] Therefore

[0345]

[0346] The solution is:

[0347]

[0348] To obtain the minimum m; simplify it to:

[0349]

[0350] Further simplified to:

[0351]

[0352] φ -1 (λ s )<0,a>b or φ -1 (λ s )>0,(φ -1 (λ s )) 2 -4ab+4a 2 >0

[0353] When a>b, there is a real number solution, and the solution that conforms to engineering practice is:

[0354] φ -1 (λ s )<0,a>b

[0355] ; Use the program algorithm to solve the delay of m cycles and draw it into a table.

[0356] Embodiment 2:

[0357] In step S8, the fault judgment switching process is as follows:

[0358] S801, the current preferred output signal is the kth path, and the first determination process and the second determination process are respectively executed;

[0359] The first judgment process is:

[0360] 11. Subtract the i-th signal from the k-th signal mean to obtain the difference, and compare the difference with βσ, i=1,2,3,…,k-1; if all the differences are greater than βσ, proceed to the next step; otherwise, return to step S801;

[0361] 12. Introduce the k-th signal deviation fault alarm and exit the alternative signal pool;

[0362] 13. Select the current preferred output signal as the k-1th channel;

[0363] 14. Subtract the i-th signal from the mean of the k-1-th signal to obtain the difference, and compare the difference with βσ, i=1,2,3,…,k-2; if all the differences are greater than βσ, proceed to the next step; otherwise, return to step 13;

[0364] 15. Let k = k-1 and return to step 12;

[0365] 16. Repeat steps 12 to 15, and complete the last cycle when k=3;

[0366] 17, output the one with the larger value between the first and second paths as the preferred output signal;

[0367] The second determination process is:

[0368] 21. Subtract the i-th signal from the mean value of the k-th signal to obtain the difference, and compare the difference with βσ, i=1,2,3,…,k-1; if all the differences are less than -βσ, proceed to the next step; otherwise, return to step S801;

[0369] 22. Introduce the k-th signal deviation fault alarm and exit the alternative signal pool;

[0370] 23. Select the current preferred output signal as the k-1th channel;

[0371] 24. Subtract the i-th signal from the mean of the k-1-th signal to obtain the difference, and compare the difference with βσ, i=1,2,3,…,k-2; if all the differences are less than -βσ, proceed to the next step; otherwise, return to step 23;

[0372] 25, let k = k-1, and return to step 22;

[0373] 26. Repeat steps 22 to 25, and complete the last cycle when k=3;

[0374] 27, output the one with the larger value among the first and second paths as the preferred output signal.

[0375] Embodiment three:

[0376] In the embodiment, the unit equipment is three sets of active signals of the hydropower unit speed governor, that is, the deviation of the active signal of the hydropower unit is determined, and the specific method is as follows:

[0377] Step 1: Establish the probability distribution of the sensor signal after filtering;

[0378] Let 1# active signal variable be ξ1, 2# active signal variable be ξ2, and 3# active signal variable be ξ3. Random variables ξ1, ξ2, ξ3 are independent of each other and obey Gaussian distribution;

[0379] Given ξ1~N(μ1,σ1 2 ),ξ2~N(μ2,σ2 2 ), After taking the mean filter, Given ε, the fault judgment of 3# active signal variable ξ3 is: When it is established, the 3# active signal is determined to be in a fault state.

[0380] in:

[0381] ξ 1i ,ξ 2i ,ξ 3i ,i=1,…,n obeys N(μ i ,σ i 2 ).

[0382] j# toothed disc mean filter signal; j=1,2,3

[0383] ξ ji : j# the i-th sampling value of the toothed disc signal;

[0384] n: The number of sampling times for signal filtering processing.

[0385] From formula (1), we get

[0386]

[0387] make

[0388]

[0389] but

[0390]

[0391] in, j# Mathematical expectation of the mean filter signal of the toothed disc; j=1,2,3; j#Standard deviation of the gear mean filter signal; j=1,2,3.

[0392] Step 2: Construct the probability distribution of signal difference;

[0393] The difference between the 3# active signal and the 1# active signal after averaging and filtering is The difference between the 3# active signal and the 2# active signal after averaging and filtering is Right now

[0394]

[0395] Knowable

[0396]

[0397] make

[0398]

[0399] in

[0400]

[0401] Column vector of the difference between the PT mean filter signal and the mean filter signals of the 1# gear disc and the 2# gear disc;

[0402] PT mean filter signal;

[0403] make

[0404] in

[0405]

[0406] Right now

[0407]

[0408] but

[0409] It follows a 3-variable normal distribution in

[0410]

[0411]

[0412] in

[0413]

[0414] Right now

[0415]

[0416] Step 3: Derive the fault recognition rate expression;

[0417] Known It follows a 3-variable normal distribution in is its subvector.

[0418] The mathematical expectation is The covariance is The following is given Under the conditions The distribution density function of .

[0419] From formula (4), we can get

[0420]

[0421] make

[0422]

[0423]

[0424] From equation (5) and equation (6), we can get:

[0425]

[0426] but

[0427]

[0428] Step 4: Derive the expression of fault misjudgment rate;

[0429] Known It follows a 3-variable normal distribution in is its subvector.

[0430] The mathematical expectation is The covariance is The following is given Under the conditions The distribution density function of .

[0431] make

[0432]

[0433] By have to

[0434]

[0435] Step 5: Calculate the confidence interval of the fault recognition rate;

[0436] Formula (12) gives have to

[0437]

[0438] make but Then from formula (7) we get

[0439]

[0440] make a≥0,b≥0, then from formula ((3), we get

[0441]

[0442] Among them, b and a are the fault threshold coefficients.

[0443] Step 6: Calculate the confidence interval of the fault misjudgment rate;

[0444] make From formula (9), we get

[0445]

[0446] From formula (10), we get

[0447]

[0448] make a≥0,b≥0, then

[0449]

[0450] Step 7: Fault diagnosis threshold setting algorithm;

[0451] According to formula (11), we can get

[0452]

[0453] Therefore

[0454]

[0455] We can get:

[0456]

[0457] The confidence interval threshold setting of fault recognition rate is given by s =0.05, k=3, the program execution scan cycle of the controller is τ=20ms, and the data collection volume of 10 years of continuous operation is:

[0458] times; calculate m by the following formula and make a table:

[0459] φ -1 (λ s )<0,a>b;

[0460] Running the algorithm program yields the following Table 1:

[0461] Table 1: Calculation results

[0462]

[0463]

[0464]

[0465]

[0466] Table 1 shows the values ​​of a, b, k, λ and m.

[0467] Embodiment 4:

[0468] The algorithm for calculating m in step S7 is:

[0469] Input: N, τ, K, T;

[0470] N: The frequency of misjudgment of equipment failure during continuous operation, i.e. once in N years;

[0471] τ: Controller program execution scan cycle;

[0472] λ s : Maximum allowable delay time for equipment fault identification;

[0473] K: number of redundant sensor channels;

[0474] a: given equipment fault judgment difference (variance multiple);

[0475] Output:

[0476] b: The variance multiple corresponding to the frequency of misjudgment of equipment faults during continuous operation; m: The equipment fault setting delay period.

[0477] Pseudocode of decision domain setting algorithm:

[0478] input N,τ,λ s ,K,a;

[0479]

[0480] Fib = 1-1 / S;

[0481] b=function_invernormcdf(Fib);

[0482]

[0483]

[0484]

Claims

1. A method for calculating the threshold value of redundancy signal deviation fault diagnosis based on joint distribution of differences and a channel switching control process thereof, characterized in that: The following steps are involved: S1, establish the probability distribution of the sensor signal after filtering; S2, construct the probability distribution of signal differences; S3, derive the expression of fault recognition rate; S4, derive the expression of fault misjudgment rate; S5, calculate the confidence interval of the fault recognition rate; S6, calculate the confidence interval of the fault misjudgment rate; S7, fault judgment domain setting value calculation: The fault misjudgment rate is set to once in N years, and the maximum allowable delay for fault identification is T; the execution cycle of the controller is τ milliseconds, and the number of data collection times S for N years of continuous operation is: Look up the normal distribution table and select the value of lower bound b>0,β=bσ, so that S(1-φ(β))×2=S(1-φ(bσ))×2<2; Set the segment delay strategy: According to the difference a, the error rate λ is given s , calculate the required delay period m, set different delay times respectively; when a>b, construct the solution equation of m, solve the delay of m periods through the program algorithm, and draw a table; S8, constructing a signal fault switching program according to the decision domain setting value obtained in step S7, and completing the equipment fault diagnosis and switching control mode.

2. The method for calculating the threshold value of the redundancy signal deviation fault diagnosis based on the joint distribution of differences and the channel switching control process thereof according to claim 1 is characterized in that: In step S1, the specific method of establishing the probability distribution of the sensor signal after filtering is: Let 1# sensor signal variable ξ1, 2# sensor signal variable ξ2, ..., k# sensor signal variable ξ k ; Random variables ξ1, ξ2, ..., ξ k They are independent of each other and obey Gaussian distribution; it is deduced that the sensor mean filter signal obeys the distribution: make but in, represents the mathematical expectation of the mean filter signal of k# sensors; represents the standard deviation of the mean filter signal of the k# sensor; μ j represents the mathematical expectation of the j# sensor filter signal; represents the mean filter signal of sensor j#; j = 1, 2, ..., k, ξ ji represents the i-th sampling value of the j# sensor signal; j=1,2,…,k, i=1,…,n; n represents the sampling times of the signal filtering process; N represents Gaussian distribution.

3. The method for calculating the threshold value of redundancy signal deviation fault diagnosis based on joint distribution of difference and the channel switching control process thereof according to claim 2 is characterized in that: In step S2, the specific method for constructing the probability distribution of the signal difference is: The difference between the k# sensor signal and the 1# sensor signal after averaging and filtering is The difference between the k# sensor signal and the k# sensor signal after averaging and filtering is The difference between the k# sensor signal and the (k-1)# sensor signal after averaging and filtering is Right now The probability distribution is make The probability distribution of the signal difference is calculated as: In the formula, is the column vector of the difference between the mean filtered signal of the k# sensor and the mean filtered signals of the 1# sensor, 2# sensor, ..., (k-1)# sensor respectively; is the mean filtered signal of the k# sensor.

4. The method for calculating the threshold value of redundancy signal deviation fault diagnosis based on joint distribution of difference and the channel switching control process thereof according to claim 3 is characterized in that: In step S3, the specific method for deriving the fault recognition rate expression is: The fault recognition rate indicates the probability that a certain signal to be detected can be effectively detected through the difference information between this signal and other signals when a deviation fault occurs. That is, when the deviation of this signal exceeds the threshold, the conditional probability that the difference between this signal and other signals exceeds the threshold is calculated as follows: Known Follows ternary normal distribution in is its subvector; The mathematical expectation is The covariance is Calculate the fault recognition rate expression:

5. The method for calculating the threshold value of redundancy signal deviation fault diagnosis based on joint distribution of difference and the channel switching control process thereof according to claim 4 is characterized in that: In step S4, the specific method for deriving the expression of fault misjudgment rate is: The fault misjudgment rate refers to the probability that the signal to be detected is normal when the system determines that the signal to be detected has a deviation fault, but in fact the signal is normal, that is, when the difference between this signal and the rest of the signals exceeds the threshold, the signal is normal, that is, the first type of error; the specific calculation process is: Known Follows k-ary normal distribution in is its subvector; the mathematical expectation is The covariance is Calculate the given Under the conditions The distribution density function is then used, and based on the fact that the variance of the same model of each sensor is the same, the fault expression is calculated as:

6. The method for calculating the threshold value of redundancy signal deviation fault diagnosis based on joint distribution of difference and the channel switching control process thereof according to claim 5 is characterized in that: In step S5, the confidence interval of the fault recognition rate is calculated, and the expression is:

7. The method for calculating the threshold value of redundancy signal deviation fault diagnosis based on joint distribution of difference and the channel switching control process thereof according to claim 6 is characterized in that: In step S6, the confidence interval of the fault misjudgment rate is calculated, and the expression is:

8. The method for calculating the threshold value of redundancy signal deviation fault diagnosis based on joint distribution of difference and the channel switching control process thereof according to claim 7 is characterized in that: In step S8, the fault judgment switching process is as follows: S801, the current preferred output signal is the kth path, and the first determination process and the second determination process are respectively executed; The first judgment process is:

11. Subtract the i-th signal from the k-th signal mean to obtain the difference, and compare the difference with βσ, i=1,2,3,…,k-1; if all the differences are greater than βσ, proceed to the next step; otherwise, return to step S801; 12. Introduce the k-th signal deviation fault alarm and exit the alternative signal pool; 13. Select the current preferred output signal as the k-1th channel; 14. Subtract the i-th signal from the mean of the k-1-th signal to obtain the difference, and compare the difference with βσ, i=1,2,3,…,k-2; if all the differences are greater than βσ, proceed to the next step; otherwise, return to step 13; 15. Let k = k-1 and return to step 12; 16. Repeat steps 12 to 15, and complete the last cycle when k=3; 17, output the one with the larger value between the first and second paths as the preferred output signal; The second determination process is:

21. Subtract the i-th signal from the mean value of the k-th signal to obtain the difference, and compare the difference with βσ, i=1,2,3,…,k-1; if all the differences are less than -βσ, proceed to the next step; otherwise, return to step S801; 22. Introduce the k-th signal deviation fault alarm and exit the alternative signal pool; 23. Select the current preferred output signal as the k-1th channel; 24, respectively, subtract the i-th signal from the mean value of the k-1-th signal to obtain the difference, and compare the difference with βσ, i = 1, 2, 3, ..., k-2; If all differences are less than -βσ, proceed to the next step; otherwise, return to step 23; 25, let k = k-1, and return to step 22; 26. Repeat steps 22 to 25, and complete the last cycle when k=3; 27, output the one with the larger value among the first and second paths as the preferred output signal.

Images