Motorized spindle reliability modeling method and system without failure data
The step-accelerated degradation test and inverse power law model generate burst-free failure data. Combined with Weibull distribution and E-bayes method, the electric spindle reliability model is constructed using IBA-SVR, which solves the problem of ignoring the impact of degradation in the non-failure data. , more accurate burst failure probability estimation and reliability modeling are achieved.
Patent Information
- Application Number
- CN202410742588.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-11
- Publication Date
- 2024-11-08
AI Technical Summary
The existing reliability modeling method of non-failure data ignores the impact of product degradation on the probability of sudden failure, resulting in high reliability estimates and failing to effectively consider the degradation information of the product during use.
Stepping accelerated degradation test and inverse power law model describe the degradation rate of the electric spindle, generate burst failure data through the maximum likelihood method, and use Weibull distribution and E-bayes method to estimate the burst failure probability, combined with the IBA-SVR method Build a reliability model that considers the degree of degradation.
The generated reliability modeling results are more realistic, taking into account the impact of the degree of degradation on the probability of burst failure, and improving the accuracy and accuracy of the model.
Smart Images

Figure CN118917167A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of reliability modeling of electric spindles, and relates to a reliability modeling method of an electric spindle taking into account the correlation between degradation degree and sudden failure under no-failure data. Background Art
[0002] The electric spindle is a high-reliability and long-life product. It rarely fails during reliability testing, and even has zero-failure data. Although zero-failure data does not contain failure data, it implies certain status information about the product life. For relatively expensive products, the problem of zero-failure data under small sample conditions has become a new field and has attracted the attention of more and more researchers. Currently, the commonly used reliability analysis methods based on zero-failure data include the confidence limit method, the distribution curve method, the equivalent failure method, etc. These methods only perform reliability modeling from the perspective of product truncated data, ignoring the degradation information during product operation, and the reliability estimation results obtained are biased high.
[0003] The performance of a product always degrades during use, and its performance degradation information can be easily obtained through sensors. According to the development process of product failure, its failure mode can be divided into two categories: degradation failure and sudden failure that occurs during the degradation process, both of which lead to product failure. Studies have shown that the probability of sudden failure is related to the degree of performance degradation. The greater the degree of performance degradation, the greater the probability of sudden failure. Therefore, when performing reliability modeling with zero-failure data, the impact of the degree of degradation on the probability of sudden failure of the product should be considered.
[0004] The above research shows that currently, reliability modeling of zero-failure data products is mainly carried out through truncated data, and the reliability modeling issues that combine degradation information during product operation and consider degradation-related issues are not involved. Summary of the invention
[0005] In view of the problem that the existing zero-failure data reliability modeling method ignores the impact of product degradation degree on the probability of sudden failure, the present invention provides a reliability modeling method for an electric spindle that considers degradation correlation under zero-failure data. The reliability modeling of the electric spindle is performed using this method, and the result is more in line with reality.
[0006] In order to solve the above technical problems, the present invention is implemented by adopting the following technical solutions, which are specifically described as follows:
[0007] A method and system for reliability modeling of an electric spindle under no-failure data, comprising the following steps:
[0008] Step 1: Step acceleration degradation test of the electric spindle and collection of degradation information;
[0009] Based on the historical fault information and load information of the same type of products, the step acceleration degradation test of the electric spindle is designed, and the step acceleration degradation test is carried out according to the design scheme to measure the degradation information of the degradation index of the electric spindle;
[0010] Step 2: Generate data on the electric spindle without sudden failure;
[0011] Assuming that the degradation process of the tested electric spindle under each step stress is a Wiener process, the inverse power law model is used to describe the relationship between the test acceleration stress and the degradation rate of the electric spindle, and the maximum likelihood method is used to obtain the degradation rate of the electric spindle under each step acceleration stress level of the electric spindle. The conversion factor is obtained by conversion, so that the step acceleration degradation test in step one is equivalent to a constant stress test, and no sudden failure data is generated;
[0012] Step 3: Estimation of the probability of sudden failure of the electric spindle;
[0013] The estimated value of the probability of sudden failure of the electric spindle at the time is obtained by using the properties of Weibull distribution, and the estimated value of the initial sudden failure probability is corrected by using the E-bayes method.
[0014] Step 4: Construction of the reliability model of the electric spindle under the no-failure data considering the degradation degree and sudden failure;
[0015] Assuming that the life of the electric spindle obeys the Weibull distribution, t~Weibull(β,η), the shape parameter β is independent of the overall degradation degree ρ(t) of the electric spindle at time t, and the scale parameter η increases with the degradation degree ρ(t), then the probability of sudden failure of the electric spindle p i =1-exp[-(T i ρ(T i ) / a) β ], by linearly fitting the degradation data, the expression of the degradation degree ρ(t) is obtained, the sudden failure probability is linearly transformed, and the Weibull distribution model parameters are obtained using the IBA-SVR method, thus obtaining the reliability model of the electric spindle considering the degradation correlation under the zero-failure data.
[0016] The step-by-step acceleration degradation test and degradation information collection of the electric spindle described in step 1 of the technical solution:
[0017] Based on the historical fault information and load information of the same type of products, the step acceleration degradation test of the electric spindle is designed, and the step acceleration degradation test is carried out according to the design scheme to measure the degradation information of the degradation index of the electric spindle;
[0018] The electric spindle has no sudden failure data generation as described in step 2 of the technical solution:
[0019] Step 1: Calculate the degradation rate of the electric spindle at each acceleration stress level;
[0020] Assuming that the degradation process of the measured electric spindle performance parameter index is a Wiener process, let X(t) represent the degradation amount of the performance parameter index at time t under a certain stress level. The degradation model of the Wiener process under this stress level is expressed as:
[0021] X(t) = x 0 +λt+σ B B(t) (1)
[0022] In the formula, x 0 is the initial degradation amount; λ is the drift coefficient, which is used to describe the degradation speed of the degradation parameter; σ B is the diffusion coefficient, which is used to describe the influence of measurement error and external disturbance random factors on the degradation process; {B(t), t≥0} is the standard Brownian motion, which reflects the randomness of the degradation process;
[0023] Assuming the failure threshold of the product is w, the life of the electric spindle T is defined according to the time when the degradation amount first reaches the failure threshold, that is:
[0024] T=inf{t:X(t)≥w|x 0 <w} (2)
[0025] Under the condition of a given drift coefficient λ, the life distribution of the electric spindle obeys the inverse Gaussian distribution:
[0026]
[0027] The inverse power law model is used to describe the influence of the accelerated stress of the step-by-step accelerated degradation test on the degradation speed of the electric spindle. For the degradation model based on the Wiener hypothesis, it is believed that the drift coefficient is highly correlated with the accelerated stress level. Taking double-step accelerated stress as an example, the inverse power law model of double-step accelerated stress is expressed as:
[0028] λ(S 1 ,S 2 )=AS 1 α S 2 β (4)
[0029] In the formula, A, α, β are constants, S 1 , S 2 are the two accelerated stresses in the test;
[0030] Diffusion coefficient σ of Wiener process under the assumption of unchanged failure mechanism B unchanged; assuming that the number of stress levels in the step accelerated degradation test is l, and the i-th step accelerated stress level is (S 1i , S 2i ), the stress application process can be expressed as:
[0031]
[0032] Where, t i represents the time after completing the i-th step stress acceleration test, t 0 =0;
[0033] According to the inverse power law model, the degradation rate under different accelerated stress levels is:
[0034] λ i (S 1 ,S 2 )=AS 1i α S 2i β (6)
[0035] The terminal value of the monitoring parameter degradation data under each accelerated stress is the initial value of the monitoring parameter degradation data under the next accelerated stress. Assuming the initial degradation amount is 0, the degradation amount of the product under the step accelerated stress is written as:
[0036]
[0037] At the i-th stress level S i =(S 1i , S 2i ) The degradation increment of the kth test is z ik , then:
[0038]
[0039] In the formula, y k is the kth test time of the product, k = ξ i-1 +1,ξ i-1 +2,...ξ i ,ξ i is the number of tests of the product when completing the i-th step acceleration, n j For products under stress S i =(S 1i , S 2i ) under the test number, if a total of G tests are performed, then G = ξ l ;
[0040] According to the properties of the Wiener random degradation process, z ik It follows a multivariate normal distribution, z ik ~N(AS 1i α S 2i β (y k -y k-1),σ B 2 (y k -y k-1 )), the maximum likelihood estimation method is used to estimate the unknown parameters, and the log-likelihood function is:
[0041]
[0042] Where N is the number of measured degradation quantities, L(·) is the likelihood function, and Z is the degradation increment z. ik A collection of;
[0043] For A, Taking partial derivatives, we can get:
[0044]
[0045] Setting the partial derivative equal to zero gives A, The maximum likelihood estimation under the conditions of α and β is shown in equations (10) and (11);
[0046]
[0047] In the formula, is the estimated value of A under the conditions of α and β;
[0048]
[0049] In the formula, is the estimated value of A under the conditions of α and β;
[0050] Substituting equations (10) and (11) into equation (9) yields the profile likelihood function for parameters α, β:
[0051]
[0052] Maximizing formula (12) can obtain the maximum likelihood estimation of α and β, which can be realized by using the FMINSEARCH function in Matlab through a multidimensional search algorithm. After obtaining the estimates of α and β, substitute the estimates of α and β into formula (10) and formula (11), and then obtain A, The maximum likelihood estimate of After estimating the value, the degradation rate of the electric spindle under different stress levels is obtained according to formula (4);
[0053] Step 2: Generate non-sudden failure data based on step-accelerated degradation test;
[0054] Divide a series of products into j groups, each with n 1 ,n 2 ,...,n gFor each group of products, a constant stress timed truncation test is carried out. The timed truncation time of each group of tests is sorted from small to large and recorded as T 1 *,T 2 *,...,T g *(T 1 *<T 2 *<...<T g *). If no product fails at the end of all tests, the data of no sudden failure is obtained, which is recorded as: (T i *,n i )(i=1,2,...,g);
[0055] Assume that the probability of product sudden failure corresponding to the test cut-off time of each group is p i , sudden failure probability p i Satisfy p 1 <p 2 <...<p g , and there are
[0056] s i =n i +n i+1 +...+n g (13)
[0057] In the formula, s i Indicates that the running time is greater than or equal to the truncation time T i *Total number of samples;
[0058] The above-mentioned no-sudden failure data is defined under the constant stress timed truncation test. Therefore, when generating no-failure data, the step acceleration degradation test should be equivalent to the constant stress test, and the step acceleration degradation test of the electric spindle should be regarded as a constant stress test of i electric spindles under i different acceleration stresses respectively; taking the degradation rate under normal working stress obtained in step 1 as the benchmark, the degradation rate under each acceleration stress level is converted to the degradation rate under normal working stress to obtain the conversion factor, and the conversion factor is multiplied by the test time under each acceleration stress level to obtain the cut-off time of each equivalent constant stress test, that is, (T i *,1)(i=1,2,...,g).
[0059] Estimation of the probability of sudden failure of the electric spindle described in step 3 of the technical solution:
[0060] Step 1: Use the properties of Weibull distribution to obtain T i *Probability of sudden failure of the electric spindle at time p i An estimated value of
[0061] When the Weibull distribution shape parameter is 0<m≤1, p iThe prior distribution of
[0062]
[0063] When the Weibull distribution shape parameter m>1, p i The prior distribution of is:
[0064]
[0065] In the formula, s i The censoring time is t i The number of samples that have not failed at the time is the upper limit of the failure probability; is the probability of sudden failure at i truncation times.
[0066] Step 2: Initial burst failure probability p based on the E-Bayes method 1 estimate revisions;
[0067] Select the E-bayes method to calculate the initial burst failure probability p 1 To make corrections:
[0068]
[0069] In the formula, c is a hyperparameter with a value range of 4 to 8. is the initial sudden failure probability p 1 The estimated value of 1 The running time is greater than or equal to the truncation time T 1 *Total number of samples;
[0070] The reliability model of the electric spindle is constructed under the non-failure data considering the degradation degree and sudden failure as described in step 4 of the technical solution:
[0071] Step 1: Construction of the reliability model of the electric spindle;
[0072] Assuming that the probability distribution of sudden failure of the electric spindle satisfies the Weibull distribution, the electric spindle fails at the i-th truncation time T i The probability of sudden failure is p i And meet:
[0073] p i =1-exp[-(T i / η) m ] (17)
[0074] The probability of sudden failure is not only related to time t, but also increases with the increase of the overall degradation degree ρ(t). Among the Weibull distribution parameters, the shape parameter m is independent of the overall degradation degree ρ(t) of the electric spindle at time t, and the scale parameter η increases with ρ(t). Assuming η = ρ(t) / a, where a is a parameter to be determined, the probability of sudden failure of the electric spindle p is i It can be expressed as:
[0075] p i =1-exp[-(T i ρ(T i ) / a) m ] (18)
[0076] Wherein, the degradation degree ρ(t) = X(t) / w, w is the failure threshold of the composite degradation, X(t) is the composite degradation of the electric spindle at time t under normal working stress, and the analytical expression of X(t) is obtained by linear fitting through the value of the composite degradation under normal working stress;
[0077] Taking the logarithm of both sides of equation (18), we can obtain:
[0078] lnln(1-p i ) -1 = mln(T i ρ(T i ))-mlna (19)
[0079] Let y = lnln(1-p i ) -1 ,φ(t)=ln(T i ρ(T i )),b=-mlna,w=m, we get:
[0080] y=wφ(t)+b (20)
[0081] Formula (20) is the same as the hyperplane f(x) = w·x + b = 0 in the support vector regression algorithm. Therefore, the estimation of parameters w and b can be realized through the SVR algorithm and converted into the estimation of Weibull distribution parameters m and η;
[0082] Step 2: Optimizing model parameters based on the IBA-SVR method;
[0083] The SVR algorithm introduces a sensitive factor ε in the loss function, indicating that the SVR algorithm can tolerate a deviation of less than ε between the model output value f(x) and the true value q. The loss calculation is performed only when |qf(x)|>ε. The loss function of the SVR algorithm is:
[0084]
[0085] Sample point (x i ,q i The distance between )∈S and the hyperplane ω·p+b=0 is:
[0086]
[0087] From formula (22), we can get that the upper bound of the distance from a point in the sample set S to the hyperplane is Maximize the upper bound of formula (22), that is, minimize ||ω|| 2 , the obtained hyperplane is the approximate optimal hyperplane, and the SVR problem is transformed into the following form:
[0088]
[0089] The hyperparameters C and ε have a great influence on the accuracy of the SVR model. It is necessary to optimize C and ε to train the optimal SVR model. The improved bat algorithm is selected to perform a global search for the hyperparameters of SVR within the specified range to ensure the accuracy of the SVR model estimation.
[0090] The Levy flight mechanism is introduced to improve the BA algorithm;
[0091] The Levy flight mechanism allows the algorithm to jump out of the local search through a variable step length, and the step length is calculated according to formula (24):
[0092]
[0093] In the formula, ρ∈[0,2], ρ=1.5; h and g are random numbers that obey the normal distribution with mean 0, and the standard deviation of the normal distribution is expressed as shown in formula (25);
[0094]
[0095] In the formula, σ h and σ g Respectively represent the standard deviation of random numbers h and g;
[0096] The pulse frequency, speed and position used by the improved BA algorithm during search are updated according to equations (26) to (28):
[0097] e i =e min +(e max -e min )·β (26)
[0098]
[0099] In the formula, e irepresents the pulse frequency emitted by bat i at the current moment; β is a random variable and takes values with equal probability in the interval [0, 1]; e max 、e min Respectively represent the maximum and minimum values of the pulse frequency; and denote the speed and position of the i-th bat at time t, respectively. represents the pulse frequency of bat i at time t, The first and second dimensions correspond to the parameters C and ε respectively; * represents the optimal position at the current moment, is the dot product operator, and Levy(δ) is the Levy vector.
[0100] A bat is randomly selected from the bat population and is asked to search locally near its current position in a random walk manner. The randomly generated new position equation is shown in equation (29).
[0101] x new =x old +υ·A t (29)
[0102] In the formula, x new Indicates the new position of the bat randomly generated, x old Indicates the current position of the bat, υ∈[-1,1], A t is the average loudness of the entire group at time t;
[0103] Bats constantly adjust the loudness and frequency of the sound waves they emit according to the location of their food; in order to quickly determine the target's location, the loudness of the sound waves increases as they gradually approach their prey. Gradually decrease, the frequency of sound waves The frequency of the sound waves of the i-th bat is increasing. and loudness Update according to formula (30) and formula (31);
[0104]
[0105] In the formula, represents the initial sound wave frequency of bat i; ψ>0 is the sound wave frequency enhancement coefficient; Q∈[0,1] is the sound wave loudness attenuation coefficient.
[0106] After the bat's position is optimized, the loudness and frequency of its sound waves will be updated, indicating that the bat is moving to the optimal position; after the bat finds its prey, it will no longer emit sound waves.
[0107] The improved BA algorithm is coupled with the SVR model. The coupled model is referred to as the IBA-SVR model. In order to obtain the optimal C and ε, the root mean square error of the model is used as the objective function of the IBA-SVR model:
[0108]
[0109] In the formula, is the objective function of the IBA-SVR model, r is the number of samples in the validation set; q i is the actual value of the i-th sample in the validation set; f i is the estimated value of the SVR model for the i-th sample in the validation set;
[0110] The SVR model parameters are obtained through the optimization problem of formula (33):
[0111]
[0112] In the formula, represents the position of the i-th bat at the k-th iteration;
[0113] After obtaining the optimal hyperparameters C and ε of SVR, the hyperplane parameters w and b are estimated by bringing the training data into the SVR model, thereby obtaining the estimated values of m and η in the Weibull distribution model of the electric spindle reliability (17).
[0114] A reliability modeling system for an electric spindle under no-failure data considering degradation degree and sudden failure, comprising:
[0115] Information collection module, used for stepping acceleration degradation test of electric spindle and degradation information collection;
[0116] A data generation module, used for generating data of the electric spindle without sudden failure;
[0117] Probability estimation module, used for estimating the probability of sudden failure of the electric spindle;
[0118] The model building module is used to build the reliability model of the electric spindle under the zero-failure data considering the degree of degradation and sudden failure.
[0119] A device, characterized in that: it includes one or more processors;
[0120] A memory for storing one or more programs;
[0121] When the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned electric spindle reliability method with zero failure data.
[0122] A computer-readable storage medium stores a computer program, wherein the program, when executed by a processor, implements the above-mentioned electric spindle reliability method under zero-failure data.
[0123] Compared with the prior art, the present invention has the following beneficial effects:
[0124] The reliability modeling method of the present invention takes into account the influence of the degradation degree of high-reliability and long-life products on the probability of sudden failure under the condition of zero failure data, and is more realistic than the traditional product reliability modeling method under zero failure data. BRIEF DESCRIPTION OF THE DRAWINGS
[0125] Figure 1 is the degradation data of the degradation performance index of the electric spindle of the present invention;
[0126] Figure 2 This is a flow chart of the electric spindle reliability modeling method described in the present invention. DETAILED DESCRIPTION
[0127] The electric spindle reliability modeling method of the present invention under the non-failure data considering the degree of degradation and the correlation between sudden failure includes the following steps: electric spindle step acceleration degradation test and degradation information collection; electric spindle non-failure data generation; electric spindle failure probability estimation; electric spindle reliability model construction under the non-failure data considering the degree of degradation and the correlation between sudden failure.
[0128] 1. Electric spindle step acceleration degradation test and degradation information collection
[0129] Based on the historical failure information and load information of the same type of products, the stepping acceleration degradation test of the electric spindle is designed, and the stepping acceleration degradation test is carried out according to the design scheme to measure the degradation information of the electric spindle degradation index.
[0130] 2. No sudden failure data generation of the electric spindle
[0131] Assuming that the degradation process of the tested electric spindle under each step stress is a Wiener process, the inverse power law model is used to describe the relationship between the test acceleration stress and the degradation rate of the electric spindle, and the maximum likelihood method is used to obtain the degradation rate of the electric spindle under each step acceleration stress level of the electric spindle. The conversion factor is obtained by conversion, so that the step acceleration degradation test in step 1 is equivalent to a constant stress test, and the data without sudden failure (T i *,n i )(i=1,2,...,j).
[0132] 3. Estimation of the probability of sudden failure of the electric spindle
[0133] Using the properties of Weibull distribution, we can get T i *Probability of sudden failure of the electric spindle at time p iThe estimated value of the initial burst failure probability p is calculated using the E-Bayes method. 1 The estimated value is corrected.
[0134] 4. Construction of the reliability model of the electric spindle under the non-failure data considering the degree of degradation and sudden failure
[0135] Assuming that the life of the electric spindle obeys the Weibull distribution, t~Weibull(β,η), the shape parameter β is independent of the overall degradation degree ρ(t) of the electric spindle at time t, and the scale parameter η increases with the degradation degree ρ(t), then the probability of sudden failure of the electric spindle p i =1-exp[-(T i ρ(T i ) / a) β ], by linearly fitting the degradation data, the expression of the degradation degree ρ(t) is obtained, the sudden failure probability is linearly transformed, and the Weibull distribution model parameters are obtained using the IBA-SVR method, thus obtaining the reliability model of the electric spindle considering the degradation correlation under the zero-failure data.
[0136] The present invention provides another embodiment, a reliability modeling system for an electric spindle under non-failure data considering the degradation degree and the correlation between sudden failure, comprising:
[0137] Information collection module, used for stepping acceleration degradation test of electric spindle and degradation information collection;
[0138] A data generation module, used for generating data of the electric spindle without sudden failure;
[0139] Probability estimation module, used for estimating the probability of sudden failure of the electric spindle;
[0140] The model building module is used to build the reliability model of the electric spindle under the zero-failure data considering the degree of degradation and sudden failure.
[0141] Based on the above-mentioned electric spindle reliability method under zero-failure data, the present invention provides another device, which includes but is not limited to: one or more processors and memories.
[0142] The memory, as a computer-readable storage medium, can be used to store software programs, computer executable programs and modules, such as program instructions corresponding to the electric spindle reliability method under no-failure data in the embodiment of the present invention. The processor executes various functional applications and data processing of the vehicle by running the software programs, instructions and modules stored in the memory, that is, realizing the above-mentioned electric spindle reliability method under no-failure data.
[0143] The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the terminal, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage devices.
[0144] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, a method for reliability of an electric spindle under no-failure data is implemented. The method for reliability of an electric spindle under no-failure data comprises the following steps:
[0145] Step 1: Step acceleration degradation test of the electric spindle and collection of degradation information;
[0146] Step 2: Generate data on the electric spindle without sudden failure;
[0147] Step 3: Estimation of the probability of sudden failure of the electric spindle;
[0148] Step 4: Construction of the reliability model of the electric spindle under the no-failure data considering the degradation degree and sudden failure;
[0149] The computer-readable storage medium provided by the present invention has computer-executable instructions which are not limited to the method operations described above, but can also execute related operations in the electric spindle reliability method under zero-failure data provided by any embodiment of the present invention.
[0150] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0151] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loading and executing computer program instructions on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be transmitted from a website site, a computer, a server or a data center to another website site, a computer, a server or a data center by wired (for example, coaxial cable, optical fiber, digital subscriber line DSL) or wireless (for example, infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server, a data center, etc. that contains one or more available media integrated. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid-state hard disk Solid State Disk, SSD), etc.
[0152] Example
[0153] Reliability modeling of electric spindles under zero-failure data considering degradation degree and sudden failure
[0154] Based on the historical failure information and load information of similar products, radial runout was selected as the performance degradation index, and the three-level dual stress step accelerated degradation test was carried out on the electric spindle. The accelerated stress type, accelerated stress level and test time at each accelerated stress level are shown in Table 1.
[0155] Table 1 Accelerated stress level of stepping acceleration degradation test of electric spindle
[0156]
[0157] Data collection is performed every 24 hours, and the degradation data obtained is normalized as follows: Figure 1 shown.
[0158] Substituting the obtained degradation data into formula (12), we can get Substituting it into equations (10) and (11), we can obtain Substituting into formula (4), we can obtain the degradation rate conversion formula under different accelerated stress levels:
[0159] λ i (S 1 ,S 2 )=(3.51E-07)×S 1i 0.986 S 2i0.734
[0160] The degradation rates under three accelerated stress levels are shown in Table 2.
[0161] Table 2 Degradation speed of electric spindle under various accelerated stress levels
[0162]
[0163] By converting the test time under each accelerated stress to the normal working condition, the cut-off time of each equivalent constant stress test can be obtained, see Table 3 for details.
[0164] Table 3. Truncation time of equivalent constant acceleration test of electric spindle
[0165]
[0166] The data of no sudden failure of the electric spindle is shown in Table 4.
[0167] Table 4 No sudden failure data of the electric spindle
[0168]
[0169] Taking c=8, the probability of sudden failure of the electric spindle at various truncation times is calculated by equations (14) and (16), as shown in Table 5.
[0170] Table 5 Probability of sudden failure of electric spindle
[0171]
[0172] The probability of sudden failure of the electric spindle p i The model is:
[0173] p i =1-exp[-(T i ρ(T i ) / a) β ]
[0174] Where ρ(t)=(8.1E-05)*t+(2.4E-04), obtained by linear fitting of the degraded data.
[0175] After the IBA-SVR algorithm, the optimal hyperparameter values of the SVR model are shown in Table 6.
[0176] Table 6 IBA-SVR parameter optimization results
[0177]
[0178] By setting the parameters in Table 6 in the SVR model, we can get b = -2.96, w = 0.321. Thus, the final reliability model of the electric spindle considering failure correlation under the condition of no sudden failure data can be obtained:
[0179]
[0180] By calculating the sum of squared errors (SSE), the validity of the reliability model of the electric spindle constructed in this section under the zero-failure data considering the degradation degree and sudden failure is verified, as shown in Table 7.
[0181] Table 7. Error sum of squares SSE
[0182]
[0183] It can be seen from Table 7 that the SSE value of the reliability model of the electric spindle under the non-failure data considering the correlation between the degree of degradation and the sudden failure is lower than that of the reliability model of the electric spindle under the non-failure data not considering the correlation between the degree of degradation and the sudden failure, indicating that the reliability model considering the correlation between the degree of degradation and the sudden failure is more consistent with the actual situation, and the SSE of the reliability model obtained by the IBA-SVR method is smaller than the SSE of the reliability model obtained by the least squares method, indicating that the reliability model of the electric spindle obtained based on the IBA-SVR method has higher accuracy, which verifies the rationality and effectiveness of this method.
[0184] Aiming at the problem that the existing zero-failure data reliability modeling method ignores the influence of product degradation degree on the probability of sudden failure, a reliability modeling method for electric spindles considering the correlation between degradation degree and sudden failure is provided. The step acceleration degradation test is equivalent to a constant addition test to generate zero-failure data, the degradation degree is integrated into the electric spindle reliability model, and the IBA-SVR method is used to estimate the parameters of the electric spindle Weibull distribution model. This method is compared with the effective electric spindle reliability model that does not consider the correlation between degradation degree and sudden failure, and it is verified that the results of this method are more in line with reality. This has certain guiding significance for product reliability modeling combined with degradation information under zero-failure data.
[0185] The above is only a specific implementation of the present invention, but the protection scope of the present invention is not limited thereto. Any modification, equivalent replacement and improvement made by any person skilled in the art within the technical scope disclosed by the present invention and within the spirit and principle of the present invention shall be covered within the protection scope of the present invention. At the same time, the contents not described in detail in this specification belong to the prior art known to those skilled in the art.
Claims
1. A reliability modeling method for an electric spindle under zero-failure data, characterized in that: The following steps are involved: Step 1: Step acceleration degradation test of the electric spindle and collection of degradation information; Based on the historical fault information and load information of the same type of products, the step acceleration degradation test of the electric spindle is designed, and the step acceleration degradation test is carried out according to the design scheme to measure the degradation information of the degradation index of the electric spindle; Step 2: Generate data on the electric spindle without sudden failure; Assuming that the degradation process of the tested electric spindle under each step stress is a Wiener process, the inverse power law model is used to describe the relationship between the test acceleration stress and the degradation rate of the electric spindle, and the maximum likelihood method is used to obtain the degradation rate of the electric spindle under each step acceleration stress level of the electric spindle. The conversion factor is obtained by conversion, so that the step acceleration degradation test in step one is equivalent to a constant stress test, and no sudden failure data is generated; Step 3: Estimation of the probability of sudden failure of the electric spindle; The estimated value of the probability of sudden failure of the electric spindle at the time is obtained by using the properties of Weibull distribution, and the estimated value of the initial sudden failure probability is corrected by using the E-bayes method. Step 4: Construction of the reliability model of the electric spindle under the no-failure data considering the degradation degree and sudden failure; Assuming that the life of the electric spindle obeys the Weibull distribution, t~Weibull(β,η), the shape parameter β is independent of the overall degradation degree ρ(t) of the electric spindle at time t, and the scale parameter η increases with the degradation degree ρ(t), then the probability of sudden failure of the electric spindle p i =1-exp[-(T i ρ(T i ) / a) β ], by linearly fitting the degradation data, the expression of the degradation degree ρ(t) is obtained, the sudden failure probability is linearly transformed, and the Weibull distribution model parameters are obtained using the IBA-SVR method, thus obtaining the reliability model of the electric spindle considering the degradation correlation under the zero-failure data.
2. The reliability modeling method of an electric spindle under zero-failure data according to claim 1 is characterized in that: The generation of the electric spindle no sudden failure data described in step 2 includes the following steps: Step 1: Calculate the degradation rate of the electric spindle at each acceleration stress level; Assuming that the degradation process of the measured electric spindle performance parameter index is a Wiener process, let X(t) represent the degradation amount of the performance parameter index at time t under a certain stress level. The degradation model of the Wiener process under this stress level is expressed as: X(t)=x0+λt+σ B B(t) (1) Where x0 is the initial degradation amount; λ is the drift coefficient, which is used to describe the degradation rate of the degradation parameter; σ B is the diffusion coefficient, which is used to describe the influence of measurement error and external disturbance random factors on the degradation process; {B(t), t≥0} is the standard Brownian motion, which reflects the randomness of the degradation process; Assuming the failure threshold of the product is w, the life of the electric spindle T is defined according to the time when the degradation amount first reaches the failure threshold, that is: T=inf{t:X(t)≥w|x0<w} (2) Under the condition of a given drift coefficient λ, the life distribution of the electric spindle obeys the inverse Gaussian distribution: The inverse power law model is used to describe the influence of the accelerated stress of the step-by-step accelerated degradation test on the degradation speed of the electric spindle. For the degradation model based on the Wiener hypothesis, it is believed that the drift coefficient is highly correlated with the accelerated stress level. Taking double-step accelerated stress as an example, the inverse power law model of double-step accelerated stress is expressed as: λ(S1,S2)=AS1 α S2 β (4) In the formula, A, α, β are constants, S1 and S2 are two accelerated stresses in the test; Diffusion coefficient σ of Wiener process under the assumption of unchanged failure mechanism B unchanged; assuming that the number of stress levels in the step accelerated degradation test is l, and the i-th step accelerated stress level is (S 1i , S 2i ), the stress application process can be expressed as: In the formula, t i It represents the time after completing the i-th step stress acceleration test, t0 = 0; According to the inverse power law model, the degradation rate under different accelerated stress levels is: λ i (S1,S2)=AS 1i α S 2i β (6) The terminal value of the monitoring parameter degradation data under each accelerated stress is the initial value of the monitoring parameter degradation data under the next accelerated stress. Assuming the initial degradation amount is 0, the degradation amount of the product under the step accelerated stress is written as: At the i-th stress level S i =(S 1i , S 2i ) The degradation increment of the kth test is z ik , then: In the formula, y k is the kth test time of the product, k = ξ i-1 +1,ξ i-1 +2,...ξ i ,ξ i is the number of tests of the product when completing the i-th step of acceleration, ξ0=0, n j For products under stress S i =(S 1i , S 2i ) under the test number, if a total of G tests are performed, then G = ξ l ; According to the properties of the Wiener random degradation process, z ik It follows a multivariate normal distribution, z ik ~N(AS 1i α S 2i β (y k -y k-1 ),σ B 2 (y k -y k-1 )), the maximum likelihood estimation method is used to estimate the unknown parameters, and the log-likelihood function is: Where N is the number of measured degradation quantities, L(·) is the likelihood function, and Z is the degradation increment z. ik A collection of; For A, Taking partial derivatives, we can get: Setting the partial derivative equal to zero gives A, The maximum likelihood estimation under the conditions of α and β is shown in equations (10) and (11); In the formula, is the estimated value of A under the conditions of α and β; In the formula, is the estimated value of A under the conditions of α and β; Substituting equations (10) and (11) into equation (9) yields the profile likelihood function for parameters α, β: Maximizing formula (12) can obtain the maximum likelihood estimation of α and β, which can be realized by using the FMINSEARCH function in Matlab through a multidimensional search algorithm. After obtaining the estimates of α and β, substitute the estimates of α and β into formula (10) and formula (11), and then obtain A, The maximum likelihood estimate of After estimating the value, the degradation rate of the electric spindle under different stress levels is obtained according to formula (4); Step 2: Generate non-sudden failure data based on step-accelerated degradation test; Divide a series of products into j groups, each group has n1, n2, ..., n g For each group of products, a constant stress timed truncation test is carried out. The timed truncation times of the tests of each group are sorted from small to large and recorded as If no product fails at the end of all tests, the data of no sudden failure is obtained, which is recorded as: (T i *,n i )(i=1,2,...,g); Assume that the probability of product sudden failure corresponding to the test cut-off time of each group is p i , sudden failure probability p i Satisfies p1<p2<...<p g , and there are s i =n i +n i+1 +...+n g (13) In the formula, s i Indicates that the running time is greater than or equal to the truncation time T i *Total number of samples; The above-mentioned no-sudden failure data is defined under the constant stress timed truncation test. Therefore, when generating no-failure data, the step acceleration degradation test should be equivalent to the constant stress test, and the step acceleration degradation test of the electric spindle should be regarded as a constant stress test of i electric spindles under i different acceleration stresses respectively; taking the degradation rate under normal working stress obtained in step 1 as the benchmark, the degradation rate under each acceleration stress level is converted to the degradation rate under normal working stress to obtain the conversion factor, and the conversion factor is multiplied by the test time under each acceleration stress level to obtain the cut-off time of each equivalent constant stress test, that is, (T i *,1)(i=1,2,...,g).
3. The reliability modeling method of an electric spindle under zero-failure data according to claim 1 is characterized in that: The estimation of the probability of sudden failure of the electric spindle described in step 3 includes the following steps: Step 1: Use the properties of Weibull distribution to obtain T i *Probability of sudden failure of the electric spindle at time p i An estimated value of When the Weibull distribution shape parameter is 0<m≤1, p i The prior distribution of When the Weibull distribution shape parameter m>1, p i The prior distribution of is: In the formula, s i The censoring time is t i The number of samples that have not failed at the time is the upper limit of the failure probability; is the probability of sudden failure at i truncation times; Step 2: Correction of the initial sudden failure probability p1 estimate based on the E-bayes method; Select the E-bayes method to correct the initial burst failure probability p1: In the formula, c is a hyperparameter with a value range of 4 to 8. is the estimated value of the initial burst failure probability p1, and s1 is the total number of samples whose running time is greater than or equal to the truncation time T1*.
4. The reliability modeling method of an electric spindle under zero-failure data according to claim 1 is characterized in that: The construction of the reliability model of the electric spindle under the non-failure data considering the degradation degree and the sudden failure mentioned in step 4 includes the following steps: Step 1: Construction of the reliability model of the electric spindle; Assuming that the probability distribution of sudden failure of the electric spindle satisfies the Weibull distribution, the electric spindle fails at the i-th truncation time T i The probability of sudden failure is p i And meet: p i =1-exp[-(T i / η) m ] (17) The probability of sudden failure is not only related to time t, but also increases with the increase of the overall degradation degree ρ(t). Among the Weibull distribution parameters, the shape parameter m is independent of the overall degradation degree ρ(t) of the electric spindle at time t, and the scale parameter η increases with ρ(t). Assuming η = ρ(t) / a, where a is a parameter to be determined, the probability of sudden failure of the electric spindle p is i It is expressed as: p i =1-exp[-(T i ρ(T i ) / a) m ] (18) Wherein, the degradation degree ρ(t) = X(t) / w, w is the failure threshold of the composite degradation, X(t) is the composite degradation of the electric spindle at time t under normal working stress, and the analytical expression of X(t) is obtained by linear fitting through the value of the composite degradation under normal working stress; Taking the logarithm of both sides of equation (18), we get: lnln(1-p i ) -1 =million(T i ρ(T i ))-million (19) Let \(y = \ln\ln(1 - p i ) -1 ,\ \varphi(t)=\ln(T i \rho(T i )),\ b = -m\ln a,\ w = m\), we get: y=wφ(t)+b (20) Formula (20) is the same as the hyperplane f(x) = w·x + b = 0 in the support vector regression algorithm. Therefore, the estimation of parameters w and b is realized through the SVR algorithm and converted into the estimation of Weibull distribution parameters m and η; Step 2: Optimizing model parameters based on the IBA-SVR method; The SVR algorithm introduces a sensitive factor ε in the loss function, indicating that the SVR algorithm can tolerate a deviation of less than ε between the model output value f(x) and the true value q. The loss calculation is performed only when |qf(x)|>ε. The loss function of the SVR algorithm is: Sample point (x i ,q i The distance between )∈S and the hyperplane ω·p+b=0 is: From formula (22), we can get that the upper bound of the distance from a point in the sample set S to the hyperplane is Maximize the upper bound of formula (22), that is, minimize ||ω|| 2 , the obtained hyperplane is the approximate optimal hyperplane, and the SVR problem is transformed into the following form: The hyperparameters C and ε have a great influence on the accuracy of the SVR model. It is necessary to optimize C and ε to train the optimal SVR model. The improved bat algorithm is selected to perform a global search for the hyperparameters of SVR within the specified range to ensure the accuracy of the SVR model estimation. The Levy flight mechanism is introduced to improve the BA algorithm; The Levy flight mechanism allows the algorithm to jump out of the local search through a variable step length, and the step length is calculated according to formula (24): In the formula, ρ∈[0,2], ρ=1.5; h and g are random numbers that obey the normal distribution with mean 0, and the standard deviation of the normal distribution is expressed as shown in formula (25); In the formula, σ h and σ g Respectively represent the standard deviation of random numbers h and g; The pulse frequency, speed and position used by the improved BA algorithm during search are updated according to equations (26) to (28): And i =and min +(and max -And min )·β (26) In the formula, e i represents the pulse frequency emitted by bat i at the current moment; β is a random variable and takes values with equal probability in the interval [0, 1]; e max 、e min Respectively represent the maximum and minimum values of the pulse frequency; and denote the speed and position of the i-th bat at time t, respectively. represents the pulse frequency of bat i at time t, The first and second dimensions correspond to the parameters C and ε respectively; * represents the optimal position at the current moment, is the dot product operator, Levy(δ) is the Levy vector; A bat is randomly selected from the bat population and is asked to search locally near the current position in a random walk manner. The randomly generated new position equation is shown in equation (29); x new =x old +υ·A t (29) In the formula, x new Indicates the new position of the bat randomly generated, x old Indicates the current position of the bat, υ∈[-1,1], A t is the average loudness of the entire group at time t; Bats constantly adjust the loudness and frequency of the sound waves they emit according to the location of their food; in order to quickly determine the target's location, the loudness of the sound waves increases as they gradually approach their prey. Gradually decreases, the sound wave frequency r i t The frequency of the sound wave of the i-th bat is r i t +1 and loudness Update according to formula (30) and formula (31); r i t+1 =r i 0 [1-exp(-ψt)] (30) In the formula, r i 0 represents the initial sound wave frequency of bat i; ψ>0 is the sound wave frequency enhancement coefficient; Q∈[0,1] is the sound wave loudness attenuation coefficient; After the bat's position is optimized, the loudness and frequency of its sound waves will be updated, indicating that the bat is moving to the optimal position; after the bat finds its prey, it will no longer emit sound waves. The improved BA algorithm is coupled with the SVR model. The coupled model is referred to as the IBA-SVR model. In order to obtain the optimal C and ε, the root mean square error of the model is used as the objective function of the IBA-SVR model: In the formula, is the objective function of the IBA-SVR model, r is the number of samples in the validation set; q i is the actual value of the i-th sample in the validation set; f i is the estimated value of the SVR model for the i-th sample in the validation set; The SVR model parameters are obtained through the optimization problem of formula (33): In the formula, represents the position of the i-th bat at the k-th iteration; After obtaining the optimal hyperparameters C and ε of SVR, the hyperplane parameters w and b are estimated by bringing the training data into the SVR model, thereby obtaining the estimated values of m and η in the Weibull distribution model of the electric spindle reliability (17).
5. A reliability modeling system for an electric spindle under no-failure data, characterized in that: include: Information collection module, used for stepping acceleration degradation test of electric spindle and degradation information collection; A data generation module, used for generating data of the electric spindle without sudden failure; Probability estimation module, used for estimating the probability of sudden failure of the electric spindle; The model building module is used to build the reliability model of the electric spindle under the zero-failure data considering the degree of degradation and sudden failure.
6. A device, characterized in that: comprising one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 4.
7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.
Citation Information
Cited By
Multi-agent collaboration-oriented block chain logistics data verification system and method
CN121563339A
