A multi-method fusion method for error elimination of full-machine fatigue test data

Abstract

The present invention belongs to the technical field of aircraft fatigue test, and specifically relates to a method for eliminating errors in the fatigue test data of the whole aircraft by integrating multiple methods. From both macroscopic and local perspectives, a technology for eliminating errors in the fatigue test data of the whole aircraft is established, which can not only identify error points that are clearly visible at the macroscopic level and improve the effect of data clustering, but also identify error points in the local data that are difficult to identify by traditional methods. Under the premise of ensuring the accuracy of error elimination, the change characteristics of the data are retained to the greatest extent. It provides high-quality data for aircraft structural damage identification based on strain data, and improves the accuracy of damage identification.

Description

Technical Field

[0001] The invention belongs to the technical field of aircraft fatigue testing, and in particular relates to a method for eliminating errors in full-aircraft fatigue test data by integrating multiple methods. Background Art

[0002] Affected by the system errors of the loading system and data acquisition system of the whole-machine fatigue test, the test site environment, and noise, the displacement and strain data of the whole-machine fatigue test will inevitably have some outliers that obviously deviate from the overall trend of the data under the same load conditions, collectively referred to as "bit errors". The existence of bit errors interferes with the structural damage identification based on strain data and reduces the accuracy of damage identification. Therefore, it is necessary to eliminate bit errors through data cleaning to avoid the above problems. This is crucial to improving the accuracy and reliability of damage identification.

[0003] During the fatigue test of the whole aircraft, when fatigue cracks are generated and extended in the key parts of the aircraft structure, the response data of the strain sensors within a certain range near the key parts show a nonlinear increase or decrease under the same load conditions, which is the key feature of damage identification. Therefore, the basic requirement of error elimination is to accurately identify the local error points and not to accidentally damage the normal data and cause the loss of data features.

[0004] In statistics, it is generally believed that system errors and noise interference obey normal distribution or Student distribution (t distribution). Under this premise, classic data error detection methods such as 3σ criterion (Laida criterion), Grubbs criterion, and Romanovsky criterion were established. However, due to the complex and diverse forms of errors in the fatigue test data of the whole machine, it was found in practical applications that these methods have a low recognition rate for local error points, and are very likely to accidentally damage normal data and cause feature loss, which does not meet the basic requirements of error elimination and reduces the sensitivity and accuracy of subsequent damage identification. Summary of the invention

[0005] The purpose of the present invention is to propose a method for eliminating data errors that meets the requirements of structural damage identification in fatigue tests of the entire aircraft, thereby avoiding the shortcomings of existing data error detection methods, improving the accuracy of error identification without losing data features, and providing a guarantee for improving the accuracy of aircraft structural damage identification.

[0006] Technical solution of the present invention: In order to achieve the above-mentioned invention object, a method for eliminating errors in full-machine fatigue test data by integrating multiple methods is proposed, which comprises the following steps:

[0007] Step 1: Extract data of several typical working conditions from the original strain data collected by all strain sensors;

[0008] Step 2: Based on the data proposed in the first step, calculate the mean s and sample standard deviation of each sensor response data under each typical working condition. Coefficient of variation δ;

[0009]

[0010] Step 3: Introduce the coefficient of variation into the likelihood calculation formula (2) to calculate the likelihood between all response data of the strain sensor and its mean under typical working conditions;

[0011]

[0012]

[0013] Where: L i represents the i-th point y in the response data of any strain gauge under a typical working condition i With the mean , s is the sample standard deviation, δ represents the coefficient of variation, and m(δ) is a function proposed by the present invention with the coefficient of variation δ as a variable.

[0014] The fourth step is to establish a likelihood-based error elimination criterion to perform preliminary error elimination on the data of each strain sensor under typical working conditions;

[0015] In the fifth step, the k-means clustering method is used to automatically divide the sensor data after macro-error elimination under typical working conditions into k clusters;

[0016] Step 6: Calculate the mean and sample standard deviation of the data in each cluster among the k clusters of sensor data under typical working conditions;

[0017] Step 7: Use the 3σ criterion to detect local data errors in each cluster and remove the detected errors until they are completely removed.

[0018] In a possible embodiment, in the first step, the typical operating condition refers to an operating condition in which the load value is between 50% and 85% of the maximum total load of the component and the frequency of occurrence in the test load spectrum is greater than 0.5 times per test flight hour.

[0019] In a possible embodiment, in the second step, the mean value of each sensor response data under each typical working condition is calculated. Sample standard deviation s, coefficient of variation δ.

[0020] In a possible embodiment, in the fourth step, the macro error elimination criterion is to set a elimination threshold value, and when the threshold value is exceeded, the macro error of the data is eliminated.

[0021] Preferably, the threshold value for macro-error elimination is a function including the coefficient of variation δ and the number of samples n1, so as to ensure that macro-errors are eliminated without losing data features and improve the universality of the method. After derivation and testing, a filtering threshold value function of formula (4) is proposed:

[0022]

[0023] When L i When ≥delta is established, y i The error point is removed.

[0024] In a possible embodiment, in the fifth step, the method for determining the k value in the k-means clustering method is as follows: according to the number of data samples n2 after eliminating macro bit errors, determine the k value range (3,3+floor(n2 / 200)), where the floor function represents rounding down; within the range, cyclically calculate the sum of squares of errors between the data mean of each cluster and the cluster center of each cluster, and the k value that minimizes the sum of squares of errors is the optimal value within the range.

[0025] In a possible embodiment, in the sixth step, for the data (y of the kth cluster), k1 ,y k2 ,…,y kn ), and use formulas (5) to (7) to calculate their means respectively Sample standard deviation k and ratio p;

[0026]

[0027]

[0028]

[0029] Preferably, if p ≥ 3, then y ki is a local bit error of the data, which is eliminated; otherwise, y ki It is normal data.

[0030] Beneficial effects of the present invention: The present invention establishes a technology for eliminating errors in fatigue test data of the entire aircraft from both macroscopic and local perspectives, which can not only identify error points that are clearly visible at a macroscopic level and improve the effect of data clustering, but also identify error points in the local data that are difficult to identify using traditional methods. Under the premise of ensuring the accuracy of error elimination, the variation characteristics of the data are retained to the greatest extent. It provides high-quality data for aircraft structural damage identification based on strain data and improves the accuracy of damage identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1The figure is a flow chart of the method of the present invention.

[0032] Figure 2 (a) is the original strain data of the strain sensor numbered 25 under typical working condition 1 in the full machine fatigue test in the preferred embodiment of the present invention.

[0033] Figure 2 (b) is the original strain data of the strain sensor numbered 210150170 under typical working condition 2 in the full machine fatigue test in the preferred embodiment of the present invention.

[0034] Figure 3 (a) is the remaining data after the error code is eliminated by using the traditional method 3σ criterion for the data of the strain sensor No. 25 in the full machine fatigue test under typical working condition 1 in the preferred embodiment of the present invention.

[0035] Figure 3 (b) is the remaining data after the error code is eliminated using the traditional method 3σ criterion for the strain sensor numbered 210150170 in the full machine fatigue test under typical working condition 2 in the preferred embodiment of the present invention.

[0036] Figure 4 (a) is the remaining data of the strain sensor No. 25 in the full machine fatigue test under typical working condition 1 in the preferred embodiment of the present invention after eliminating the macro bit errors, and the clustering result of the remaining data. The black circles in the figure represent the clustering center of each cluster.

[0037] Figure 4 (b) is the clustered data of the strain sensor No. 25 in the full machine fatigue test under typical working condition 1 in the preferred embodiment of the present invention after eliminating the macro bit errors, and the remaining data after eliminating the local bit errors in each cluster using the 3σ criterion.

[0038] Figure 5 (a) is the remaining data of the strain sensor numbered 210150170 in the full machine fatigue test under typical working condition 2 in the preferred embodiment of the present invention after eliminating the macro bit errors, and the clustering result of the remaining data. The black circles in the figure represent the clustering center of each cluster.

[0039] Figure 5 (b) is the clustered data of the strain sensor numbered 210150170 in the full machine fatigue test under typical working condition 2 in the preferred embodiment of the present invention after eliminating the macro bit errors, and the remaining data after eliminating the local bit errors in each cluster using the 3σ criterion. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] like Figure 1 As shown, a multi-method fusion method for eliminating errors in full-machine fatigue test data includes the following steps:

[0042] The first step is to statistically analyze the test load spectrum before the test, and select two typical load conditions from the design load spectrum, which are recorded as typical conditions 1 and typical conditions 2, respectively. Their occurrence frequencies are 1.442 times / test flight hour and 0.825 times / test flight hour, respectively, and their wing root bending moments are 66.1% and 82.6% of the maximum wing root bending moment in the load spectrum, respectively. From the original data of the full-aircraft fatigue test, the original strain data of each sensor under the typical conditions 1 and typical conditions 2 are extracted, including the data of strain sensors numbered 25 and 210150170.

[0043] The second step is to calculate the mean value of the data of the strain sensor numbered 25 under typical working condition 1. Sample standard deviation s = 258.6, coefficient of variation δ = 0.1207, sample number n1 = 19923; the mean value of the data of the strain sensor numbered 210150170 under typical working condition 2 is calculated The sample standard deviation is s=45.6, the coefficient of variation is δ=0.0192, and the sample number is n1=2934.

[0044] Step 3: Use the likelihood calculation formula (2) to calculate the likelihood between each data of the strain sensor numbered 25 and its mean under typical working condition 1, where m(δ)=m(0.1207)=0.9985; calculate the likelihood between each data of the strain sensor numbered 210150170 and its mean under typical working condition 1, where m(δ)=m(0.1207)=0.9985; calculate the likelihood between each data of the strain sensor numbered 210150170 and its mean under typical working condition 2, where m(δ)=m(0.1207)=0.9985; calculate the likelihood between each data of the strain sensor numbered 210150170 and its mean under typical working condition 1, where m(δ)=m(0.1207)=1.0.

[0045] Step 4: Establish a likelihood-based error rejection criterion. Using formula (4), calculate the macro error rejection threshold value delta=374.6147 for strain sensor No. 25 under typical working condition 1. Determine whether the likelihood of each data point and the mean is ≥ delta. If it is greater, the sample point is an error and is rejected. Repeat steps 2 to 4 until no sample points that meet the error rejection criterion are found. The remaining data after removing the macro error is as follows: Figure 4 (a) shown.

[0046] Similarly, the macro error elimination threshold value delta=82.9625 of the strain sensor numbered 210150170 under typical working condition 2 is calculated. The remaining data after eliminating the macro error is as follows: Figure 5 (a) shown.

[0047] In the fifth step, the k-means clustering method is used to automatically divide the remaining data of the strain sensor numbered 25 in typical working condition 1 into k clusters. According to the number of data samples n2=19617 after eliminating the macro bit error, the value interval of k is determined to be (3,3+floor(n2 / 200)). After iterative calculation, when k=6, the sum of square errors between the mean of each cluster and the cluster center is the smallest. Therefore, the data is divided into 6 clusters. The results are as follows: Figure 4 As shown in (a), the location of the circle in the figure represents the clustering center of each cluster.

[0048] Similarly, the remaining data of the strain sensor numbered 210150170 is automatically divided into 4 clusters. According to the number of data samples after removing the macro bit error, n2 = 2853, the value interval of k is determined to be (3, 3 + floor (n2 / 200)) = (3, 17). After iterative calculation, when k = 7, the sum of squares of the errors between the mean of each cluster and the cluster center is the smallest. Therefore, the data is divided into 7 clusters. The results are as follows: Figure 4 As shown in (a), the location of the circle in the figure represents the clustering center of each cluster.

[0049] Step 6: For the six clusters of strain sensors numbered 25 and the seven clusters of strain sensors numbered 210150170, use formulas (5) to (7) to calculate the data mean of each cluster: Sample standard deviation k and ratio p;

[0050] Step 7: According to the 3σ criterion, if p ≥ 3, the corresponding sample point is an error point, and it is removed from the corresponding cluster. After the 3σ criterion is used to remove the error in the 6 clusters of the strain sensor numbered 25, the remaining normal data is as follows: Figure 4 (b) shows that after the 3σ criterion is used to eliminate the bit errors in the 7 clusters of the strain sensor numbered 210150170, the remaining normal data is as follows: Figure 5 (b) as shown.

[0051] In the above specific embodiment, the effect of using the method of the present invention to remove data errors in the whole machine fatigue test is shown in Figure 4 , Figure 5 For the same data, if the 3σ criterion is directly used for error elimination, the final error elimination effects of the data of the strain sensor numbered 25 and the data of the strain sensor numbered 210150170 are shown in Figure 2. Figure 3 (a) Figure 3 (b) By comparison, it can be seen that the bit errors are not completely eliminated by directly using the 3σ criterion. Figure 3 (a)) or caused significant feature loss ( Figure 3 (b)), and the method of the present invention avoids these problems. It provides high-quality data for further aircraft structure damage identification based on strain data, which is conducive to improving the accuracy of damage identification.

[0052] The above is only a specific embodiment of the present invention, and the present invention is described in detail. The unexplained part is a conventional technology. However, the protection scope of the present invention is not limited to this. Any changes or substitutions that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A multi-method fusion method for eliminating errors in full-machine fatigue test data, characterized in that: The steps include: Step 1: Extract data of several typical working conditions from the original strain data collected by all strain sensors; Step 2: Based on the data proposed in the first step, calculate the mean, sample standard deviation, and coefficient of variation of each sensor response data under each typical working condition; Among them, the mean Sample standard deviation s, coefficient of variation δ; Step 3: Introduce the coefficient of variation into the likelihood calculation formula (2) to calculate the likelihood between all response data of the strain sensor and its mean under typical working conditions; Where: L i represents the i-th point y in the response data of any strain gauge under a typical working condition i With the mean , s is the sample standard deviation, δ represents the coefficient of variation, and m(δ) is a function proposed by the present invention with the coefficient of variation δ as a variable; The fourth step is to establish a macroscopic error elimination criterion based on likelihood and perform preliminary error elimination on the data of each strain sensor under typical working conditions; In the fifth step, the k-means clustering method is used to automatically divide the sensor data after preliminary error elimination under typical working conditions into k clusters; Step 6: Calculate the mean and sample standard deviation of the data in each cluster among the k clusters of sensor data under typical working conditions; Step 7: Use the 3σ criterion to cyclically detect data errors in each cluster, and remove the detected errors until the errors are completely removed.

2. The method for eliminating errors in full-machine fatigue test data by integrating multiple methods according to claim 1 is characterized in that: In the first step, the typical operating condition refers to an operating condition in which the load value is between 50% and 85% of the maximum total load of the component and the frequency of occurrence in the test load spectrum is greater than 0.5 times per test flight hour.

3. The method for eliminating errors in full-machine fatigue test data by integrating multiple methods according to claim 1 is characterized in that: In the fourth step, the macro error elimination criterion is to set a elimination threshold value, and when the threshold value is exceeded, the data is eliminated.

4. The method for eliminating errors in full-machine fatigue test data by integrating multiple methods according to claim 3 is characterized in that: The threshold value for macro-error elimination is a function of the coefficient of variation δ, the sample standard deviation s, and the number of samples n1, ensuring that macro-errors are eliminated without losing data features. The elimination threshold value function of formula (4) is proposed: Where n is the number of data samples of a single sensor under a typical working condition, and min(m(δ)) is the minimum value of min(m(δ)).

5. The method for eliminating errors in full-machine fatigue test data by integrating multiple methods according to claim 4 is characterized in that: When L i When ≥delta holds true, the corresponding y i The error point.

6. The method for eliminating errors in full-machine fatigue test data by integrating multiple methods according to claim 5 is characterized in that: In the fifth step, the method for determining the optimal value of k in the k-means clustering method is as follows: according to the number of data samples n2 after eliminating macro bit errors, determine the value interval of k (3,3+floor(n2 / 100)), the floor function represents rounding down, and cyclically calculate the sum of squares of errors between the data mean of each cluster and the cluster center of each cluster within the interval. The k value with the smallest sum of squares of errors is the optimal value in the interval.

7. The method for eliminating errors in full-machine fatigue test data by integrating multiple methods according to claim 6 is characterized in that: In the sixth step, for the data of the kth cluster (y k1 ,y k2 ,…,y kn ), and use formulas (5) to (7) to calculate their means respectively Sample standard deviation k and ratio p; 。 8. The method for eliminating errors in full-machine fatigue test data by integrating multiple methods according to claim 7 is characterized in that: In the seventh step, if p ≥ 3, then y ki is a local bit error of the data, which is eliminated; otherwise, y ki It is normal data.

Images