Spaceflight sealed relay redundancy detection signal classification method, system and clustering method
Through the method of pulse extraction, frame division, folding and k-CAD clustering combined with AE-LOF algorithm, the problem of low accuracy in the detection signal classification of the aerospace relay redundant is solved, and high-accuracy signal recognition is achieved, and the detection results are improved. reliability.
Patent Information
- Application Number
- CN202411234858.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-04
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-09-04
AI Technical Summary
The existing classification methods for detecting redundant signals of aerospace relays are difficult to accurately identify redundant signals and component signals, resulting in low detection accuracy, especially in the case of signal aliasing.
Pulse extraction, framing, folding and k-CAD clustering methods are used, and anomaly evaluation is performed in combination with the AE-LOF algorithm to realize the classification of redundant detection signals. Specific steps include pulse extraction, frame division, folding, clustering and anomaly evaluation, clustering and outlier detection of data points through peak points on the x-y coordinate axis, and then signal classification.
The accuracy of identification of excess detection signals of aerospace seal relays has been significantly improved, including six types of accurate recognition rates, including single-component signals, mixed signals, and excess signals, reaching 96.50% to 98%.
Smart Images

Figure CN119066473A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of aerospace relay redundant object detection, and in particular relates to an aerospace relay redundant object detection signal classification method, system and redundant object detection signal clustering method. Background Art
[0002] As a switching component, aerospace relays are an extremely important part of the spacecraft system. Before becoming a qualified product, aerospace relays need to undergo multiple mandatory tests, and redundant object detection is an important part of the mandatory test. Studies have shown that the redundant object problem is a serious problem faced by aerospace products. Based on existing relevant data, it is estimated that redundant objects account for about 6% of the total number of aerospace system failures. A large number of cases have shown that the redundant object problem cannot be completely avoided, so redundant object detection is currently the only economical and effective solution. Particle noise collision detection (hereinafter referred to as: PIND) equipment is a special equipment for redundant particle detection of sealed components. Its purpose is to effectively detect loose particles in components and semiconductor packaging cavities, thereby improving the reliability of components, semiconductors and other components. Particle noise collision detection method (hereinafter referred to as: PIND method) is a standard method for redundant object detection. The accuracy of the detection results of the particle noise collision detection system is extremely important. Similarly, the accuracy of the redundant object detection results of aerospace relays based on the PIND method is extremely important for the quality and reliability of aerospace relays.
[0003] At present, one of the difficulties in judging the results of aerospace relay redundant detection is the accurate identification of the redundant collision signal inside the tested piece and the component mechanical vibration signal (referred to as component signal). It should be noted that in many cases, these two signals will appear at the same time and mix together to form a mixed signal. In mixed signals, how to effectively identify the redundant signal has always been a difficult problem in this field. Since it is not easy to accurately identify the redundant signal, it is often considered to give priority to identifying the component signal. The currently widely used component signal identification method is the pulse occurrence time method, that is, it is assumed that the component pulse always appears in equal periods. However, based on the long-term aerospace relay redundant detection results and analysis, it is shown that the component signal is not always a single group of pulse sequences that appear continuously and in equal periods. This feature leads to many misjudgments and missed judgments based on the pulse occurrence time method. It is precisely because of this problem that the existing particle noise collision detection method still has a lot of room for improvement in the discrimination method.
[0004] In addition, Figure 1 As shown, the motion trajectory of the redundant objects is usually random, and the motion trajectory of the components is periodic. Figure 1 (a) shows the motion trajectory of the excess material and components under periodic vibration. Figure 1(b) shows the pulses generated by the collision between the excess particles and the components at different times. Figure 1 In (a), t is the time axis, S is the displacement axis, H is the height of the cavity of the sealed relay moving with constant amplitude in a single cycle, t k (k=1,…,∞) is the collision time of the excess particles, t m (m=1,…,∞) is the moment of component collision, and the motion trajectories of the upper and lower walls of the cavity are represented by yellow and green lines respectively, which shows the motion state of the cavity. Figure 1 In (b), the reference period is used to illustrate the periodic nature of the vibration. Figure 1 The red trajectory in (a) describes the motion of the excess in the periodic vibration system. The excess repeatedly collides between the upper and lower walls, and its path is irregular and difficult to predict. Figure 1 The expression forms of the redundant object signals and component signals obtained by the PIND method can be better understood. In fact, the redundant object detection signals of aerospace sealed relays obtained based on the PIND method present various types. However, previous studies have simply classified the loose particle detection signals in the reliability assessment of aviation sealed relays, and the analysis of the waveform expression forms is not comprehensive. Moreover, it is often difficult to accurately identify multiple types based on the static characteristics of the signal, so the accuracy of redundant object detection is low. In order to solve these problems, it is necessary to introduce a more reasonable and systematic signal processing method, especially in the face of the aliasing problem of non-stationary redundant object signals and stable component signals in the redundant object detection signal. Therefore, the existing aerospace sealed relay redundant object detection signal recognition method still has a relatively large room for improvement in the recognition effect of redundant objects. In order to ensure the high safety and high stability requirements of spacecraft, for aerospace relays, the final recognition accuracy needs to be further improved. Summary of the invention
[0005] The present invention aims to solve the problem that the classification accuracy of aerospace relay redundant object detection signals needs to be further improved.
[0006] A method for classifying redundant object detection signals of an aerospace sealed relay comprises the following steps:
[0007] Pulse extraction is performed on the PIND signal collected in the detection of redundant objects in aerospace relays, and the extracted pulse signal is framed; then the framed signal is folded;
[0008] The pulse is represented by the coordinates of the peak point of each pulse in the sub-frame where the pulse is located, and the peak points of the pulses on all sub-frame segments are displayed on the xy coordinate axis, where the x-axis is the length direction of each frame signal and the y-axis is the amplitude direction of each frame signal;
[0009] Based on the peak point coordinates of the pulses corresponding to all the frame segments on the xy coordinate axis, multiple data points in the xy coordinate system are obtained. For the multiple data points, k-CAD is used for clustering. At the same time, the abnormality of the outliers is evaluated based on the peak point coordinates of the pulses corresponding to all the frame segments on the xy coordinate axis. Then, the detection signal is classified based on the clustering results and the abnormality.
[0010] Furthermore, in the process of framing the extracted pulse signal, the framing is performed using the vibration cycle time length of the vibration table as the length of the framing window.
[0011] Furthermore, before the pulse signal is framed, adjacent pulses whose time interval is less than a set threshold are combined into one pulse.
[0012] Furthermore, for multiple data points in the xy coordinate system, the process of clustering using k-CAD includes the following steps:
[0013] The set of data points corresponding to the peak points of the pulses corresponding to all segmented segments on the xy coordinate axis is recorded as data set C:
[0014] C={x i |i=1,2,…,n},x i =[x i1 ,x i2 ,…,x im ]
[0015] Among them, x i represents a data point sample, [x i1 ,x i2 ,…,x im ] is x i The coordinates of , n is the number of data point samples;
[0016] Based on the optimal number of clusters k, the k-means clustering algorithm is used to cluster the data set C to obtain the category label L of each point = [l 1 ,l 2 ,…,l n ], where l i Represents the signal point x i Category, l i ∈{1,2,…,k}; for each category j, j=1,2,…,k, extract all sample points belonging to this category, recorded as set C j ; The center μ of category j j The calculation formula is:
[0017]
[0018] Among them, |Cj | is the number of sample points of category j;
[0019] The k-CAD clustering result of category j is
[0020] Furthermore, the optimal number of clusters k is determined according to the following steps:
[0021] The k-means clustering algorithm is used to divide the data set C into k categories. The SSE under different k values is calculated, and the relationship between the k value and the sum of squared errors within the cluster is plotted; the k value at the inflection point is used as the final determination of the optimal number of clusters.
[0022] Furthermore, the AE-LOF algorithm is used to evaluate the abnormality of the outliers based on the peak point coordinates of the pulses corresponding to all the frame segments on the xy coordinate axis. The specific process includes:
[0023] For the peak point coordinates of the pulses corresponding to all the frame segments on the xy coordinate axis, the associated point set A is first obtained. i ; For each data point p i , the set of associated points A i as follows:
[0024] A i ={p j ∈C\{p i}|d(p i , p j )≤R i}
[0025] This set contains all the i The radius R i Other data points within d k (·) represents the k-neighbor distance;
[0026] For each point p in the data set j , initialize a frequency counter F j =0; for each data point p i , check every other point p j Is it in A i If in then F j =F j +1; After traversing all points, each F j will represent the point p j The total number of times the associated point appears within the radius of all other points;
[0027] Then calculate AE-LOF(p j ):
[0028]
[0029] Among them, N k (·) represents the set of neighbor points, |N k (P)| represents the set capacity; LRD(·) represents the inverse of the average reachable distance from a point to its surrounding neighboring points; Z is the internal adjustment coefficient, and w=Z×k is the external adjustment coefficient;
[0030] Using AE-LOF(p j ) evaluates the abnormality of outliers based on the peak points of the corresponding pulses on all frame segments on the xy coordinate axis.
[0031] Furthermore, the process of classifying the detection signal based on the clustering result and the abnormality degree includes the following steps:
[0032] For multiple data points in the xy coordinate system, calculate the pulse duty cycle D of the pulse corresponding to each point. When D ≥ the duty cycle threshold, it is identified as an excessive component signal; otherwise, continue to identify;
[0033] Then calculate all the points Among them, |C| represents the capacity of the data set C corresponding to all points, μ j Represents the C mean of the data set; when CAD(xi)≤clustering threshold, and the AE-LOF algorithm detection result has no outliers, the identification result is a single-component signal; when CAD(xi)≤clustering threshold, and the AE-LOF algorithm detection result has outliers, the identification result is a mixture of a single-component signal and a redundant particle signal; otherwise, continue to identify;
[0034] The calculation result of the k-CAD algorithm corresponding to the optimal number of clusters k is recorded as CAD k (xi) If CAD k (xi)≤clustering threshold, and the AE-LOF algorithm detection result has no outliers, and the identification result is a multi-component signal; if CAD k (xi)≤clustering threshold, and the AE-LOF algorithm detection result has an outlier, and the identification result is a mixture of multi-component signals and redundant particle signals; if CAD k (xi)>clustering threshold, and the AE-LOF algorithm detection result has an abnormal value, and the identification result is an extraneous particle signal.
[0035] An aerospace sealed relay redundant object detection signal classification system, comprising:
[0036] Pulse extraction module: extract pulses from PIND signals collected during redundant object detection of aerospace relays;
[0037] Pulse segmentation module: divide the extracted pulse signal into frames;
[0038] Signal folding and coordinate projection module: fold the framed signal; then use the coordinates of the peak point of each pulse in the frame where the pulse is located to represent the pulse, and display the peak points of the pulses on all frame segments on the xy coordinate axis, where the x-axis is the length direction of each frame signal and the y-axis is the amplitude direction of each frame signal;
[0039] Classification module of detection signal: Based on the peak point coordinates of the pulses corresponding to all the frame segments on the xy coordinate axis, multiple data points in the xy coordinate system are obtained. For multiple data points, k-CAD is used to perform clustering. At the same time, the abnormality of outliers is evaluated based on the peak point coordinates of the pulses corresponding to all the frame segments on the xy coordinate axis. Then, the detection signal is classified based on the clustering results and the abnormality.
[0040] Furthermore, the classification module of the detection signal includes a clustering result calculation unit, an outlier abnormality evaluation unit, and a classification unit; wherein,
[0041] Clustering result calculation unit: based on the peak point coordinates of the pulses corresponding to all the frame segments on the xy coordinate axis, multiple data points in the xy coordinate system are obtained, the CAD values of all the points are calculated for the multiple data points, and at the same time, based on the optimal clustering number k, the k-CAD is used to cluster the multiple data points to obtain the k-CAD value;
[0042] Outlier abnormality degree assessment unit: assesses the abnormality degree of the outlier based on the peak point coordinates of the corresponding pulses on all frame segments on the xy coordinate axis;
[0043] Classification and identification unit: classifies the detection signal according to the result of the clustering result calculation unit and the result of the outlier abnormality evaluation unit.
[0044] A method for clustering redundant object detection signals for classifying redundant object detection signals of aerospace sealed relays comprises the following steps:
[0045] Pulse extraction is performed on the PIND signal collected in the detection of redundant objects in aerospace relays, and the extracted pulse signal is framed; then each framed signal is used as a data point sample x i , convert each framed signal into an m-dimensional vector [x i1 ,x i2 ,…,x im ], use [x i1 ,x i2 ,…,x im ] as the data point x i The coordinates of ; all sets of data points are recorded as data set C:
[0046] C={x i |i=1,2,…,n},x i =[x i1 ,x i2 ,…,x im ]
[0047] Among them, x i represents a data point sample, [x i1 ,x i2 ,…,x im ] is x i The coordinates of , n is the number of data point samples;
[0048] Based on the optimal number of clusters k, the k-means clustering algorithm is used to cluster the data set C to obtain the category label L of each point = [l 1 ,l 2 ,…,l n ], where l i Represents the signal point x i Category, l i ∈{1,2,…,k}; for each category j, j=1,2,…,k, extract all sample points belonging to this category, recorded as set C j ; The center μ of category j j The calculation formula is:
[0049]
[0050] Among them, |C j | is the number of sample points of category j;
[0051] The k-CAD clustering result of category j is
[0052] k-CAD(C j ) is used for the classification of redundant object detection signals of aerospace sealed relays.
[0053] Beneficial effects:
[0054] The present invention can effectively improve the recognition accuracy of redundant object detection signals for different types of aerospace sealed relays, and can identify six types of signal confusion, namely, single-component signal, mixed signal I, redundant object signal, mixed signal II, multi-component signal and oversized component signal. The accuracy of the single-component signal is 96.50%, the accuracy of the mixed signal I is 95.00%, the accuracy of the redundant object signal is 96.00%, the accuracy of the mixed signal II is 93.00%, the accuracy of the multi-component signal is 92.00%, and the accuracy of the oversized component signal is 98%. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] Figure 1Schematic diagram of the motion trajectory and collision pulse of excess particles and components under periodic vibration.
[0056] Figure 2 Schematic diagram of pulse extraction and waveform segmentation process;
[0057] Figure 3 Schematic diagram of pulse framing and folding steps of redundant signal;
[0058] Figure 4 It is the waveform diagram of the original component signal of the mixed signal;
[0059] Figure 5 This is a pulse point display diagram after mixed signal processing;
[0060] Figure 6 Schematic diagram of outliers and clusters;
[0061] Figure 7 Schematic diagram of the local outlier detection algorithm;
[0062] Figure 8 This is a schematic diagram of the degree of abnormality of outliers based on the LOF algorithm;
[0063] Fig. 9 It is a schematic diagram of the degree of abnormality of outliers based on the AE-LOF algorithm;
[0064] Fig.10 It is a comparison diagram of the LOF value and AE-LOF value of a certain mixed signal data;
[0065] Fig.11 It is a schematic diagram of the before and after calculation values based on the k-cluster center average distance method;
[0066] Fig.12 A schematic diagram of determining the number of clusters k after clustering based on the elbow method;
[0067] Fig.13 is a flow chart of the classification system of the present invention;
[0068] Fig.14 This is the recognition effect diagram of a single component signal;
[0069] Fig.15 This is the identification effect diagram of the redundant particle signal;
[0070] Fig.16 This is the identification effect diagram of the mixed signal I, which is a mixture of a single component signal and a redundant particle signal;
[0071] Fig.17 This is the identification effect diagram of mixed signal II, which is a mixture of multi-component signals and redundant particle signals;
[0072] Fig.18This is a diagram showing the effect of multi-component signal identification;
[0073] Fig.19 A line chart comparing the accuracy of the new and old identification methods on different signal types. DETAILED DESCRIPTION Specific implementation method one:
[0075] This embodiment is a method for classifying aerospace sealed relay excess detection signals, comprising the following steps:
[0076] Step 1: PIND signal preprocessing:
[0077] S100, extract pulses from the PIND signal. In this embodiment, a three-threshold method is used to extract valid pulses from the signal. It should be noted that the present invention includes but is not limited to the three-threshold method for extracting valid pulses from the signal, and may also be a double-threshold method, etc., as long as valid pulses from the signal can be extracted.
[0078] The process of extracting effective pulses from the signal using the three-threshold method is as follows:
[0079] (a) Determination of reference threshold: First, the overall acquisition signal is taken as the object, and the average energy is calculated as the threshold reference. The pulse main energy threshold, the starting energy threshold and the ending energy threshold are determined according to the average energy, such as Figure 2 (a) The threshold parameters need to be adjusted according to the actual situation and signal characteristics.
[0080] (b) Preliminary search: According to the principles of the search algorithm, a single pulse signal is processed, such as Figure 2 (b) is shown. Determine whether there is a pulse that exceeds the threshold. When the average energy value or amplitude of a certain segment of signal data reaches the set threshold, it is determined that the pulse body exists. Then, continue to calculate the energy of each segment backward until the energy value is lower than the set main energy threshold. Count the energy value peak point and obtain the time starting point and position coordinates of the segment with the highest energy.
[0081] (c) Detailed search: Based on the preliminary search, the pulse energy peak point is used as the starting point, and the set time length is used as the step length to search forward and backward for the exact starting point and the exact end point of the pulse occurrence time. Figure 2 (c) as shown.
[0082] At this point, the extraction of the pulses in the signal is completed, ensuring the accuracy and integrity of the pulses.
[0083] S200, pulse merging, segmentation, folding:
[0084] For the extracted pulse signal, the vibration period of the vibration table is used as the length of the frame window (i.e., the segmentation step) for framing, that is, segmentation is performed according to the segmentation step, and then the segmented signal is folded, such as Figure 2 As shown in (d) in the figure. Before segmentation, the acoustic reflection characteristics and sensor sensitivity characteristics are considered, and adjacent pulses with a time interval less than a set value are merged into one pulse, that is, adjacent pulses that are too close are merged.
[0085] (1) The segmentation is performed using the period of the vibration periodic signal as the step length (or the length of the divisible frame window).
[0086] (2) The merging of adjacent pulses is completed before the segmentation process is carried out, that is, many pulses are screened out through the three-threshold method, such as pulse a1, pulse a2, pulse a3, ..., pulse an obtained from a certain signal. Among them, pulses a1 and a2 are very close, and the distance between the end point of pulse a1 and the starting point of pulse a2 is less than the set value Δdmin=500, then a1 and a2 are merged, that is, the end point of pulse a2 is assigned to the end point of pulse a1. Therefore, the actual range of the merged pulse a1-2 is: [starting point of a1, end point of a2].
[0087] (3) The folding process is carried out after the segmentation process, such as Figure 3 Specifically, the frame data obtained after the segmentation program is executed are parallelized. Figure 3 Finally, the peak point of the pulse in each sub-frame is extracted as the representative of the pulse and displayed to form Figure 3 The final two-dimensional scatter plot in .
[0088] In fact, the x-axis is the length direction of each frame signal, the y-axis is the amplitude direction of each frame signal, and the z-axis is the superposition direction of each frame signal. In the actual drawing, the range on the x-axis is the vibration cycle time multiplied by the sampling rate of the detection system (500k in this embodiment), which is used to represent the data position point.
[0089] S300, pulse display: The pulse is represented by the coordinates of the peak point of each pulse in the subframe where the pulse is located, and the peak points of the pulses on all the segmented segments are displayed on the xy coordinate axis. Figure 2 (e) shows the coordinates of the pulse peak point after folding.
[0090] Step 2: Pulse point display after mixed signal preprocessing:
[0091] Macro waveform of mixed signal in aerospace relay redundant object detection and pulse peak point distribution image representation after algorithm processing:
[0092] (1) Affected by objective factors such as the shape, mass, material, and collision angle of the unwanted particles, the trajectory of the unwanted signal has a randomness similar to that of the free electron collision signal, and the peak point image of the unwanted signal pulse is an irregular scatter plot.
[0093] (2) The pulse peak point image of the component signal has the contraction of the cluster data set and looks like a narrow band of cloud.
[0094] (3) Figure 4 The figure shows the detection signal diagram of the mixed signal. It is difficult to see the regularity from the detection signal diagram. In fact, the pulse peak point distribution image of the mixed signal has two characteristics: (1) The main part of the pulse peak points is distributed in a cluster shaped like a narrow band cloud, and most of them are distributed inside and around the cluster, such as Figure 5 The horizontal coordinate is 0.4×10 4 (2) A small portion of the pulse peak points are randomly distributed in other areas, like scattered stars.
[0095] In fact, the pulse point distribution after mixed signal preprocessing reflects both the periodicity of component signals and the randomness of redundant signals. Figure 5 It can be seen that the main problem of the present invention is to determine whether there are redundant signals shaped like stars outside the component signal shaped like a narrow band cloud. It can be further understood that whether there are redundant particle signals around the component signal with good periodicity is detected.
[0096] It needs to be further explained that the component signal that resembles a narrow band cloud has a certain range of mobile redundancy, and only the pulse peak point of the redundant signal that is far away from the "narrow band cloud" needs to be clearly identified.
[0097] Step 3: For all points on the xy coordinate axis, use the k-CAD algorithm to cluster the data points, evaluate the abnormality based on the clustering results and the AE-LOF algorithm, and then use the classification system to classify the detection signal. The AE-LOF algorithm and the k-CAD algorithm are described below.
[0098] (A) The AE-LOF algorithm is used to evaluate the abnormality degree of the peak point coordinates of the pulses corresponding to all segmented segments on the xy coordinate axis.
[0099] LOF algorithm principle (conventional existing algorithm):
[0100] In 2000, Markus M.Breunig et al. proposed the Local Outlier Factor (LOF) method. This method is used to identify abnormal data. Its core idea is to determine whether a point is abnormal by comparing the local density of a point with its neighboring points. Therefore, compared with the distance-based anomaly algorithm, the density-based outlier detection can detect a type of abnormal data - local outliers. In the LOF algorithm, the degree of abnormality of the data is defined based on the concept of relative density, so it has a strong dependence on the surrounding data.
[0101] Figure 6 It is a two-dimensional data set, and the graph contains two clusters C 1 ={c 11 ,c 12 ,…,c 1n} and C 2 ={c 21 ,c 22 ,…,c 2n}, two types of outliers {o 1 ,o 3} and {o 2 ,o 4}. C 1 is a dense morphological cluster, C 2 is a sparse morphological cluster. 1 ,o 3} is a global outlier, {o 2 ,o 4} is a local outlier. Generally speaking, global outliers are easy to identify, but local outliers are difficult to mine. If you want to obtain local outliers, you need to adjust the distance parameter Dist(x). If Dist(x) is smaller than C 1 With o 2 The minimum distance between them, then C 2 Some data points in {c 2i ,c 2j} may be mistakenly identified as an outlier. Based on the above analysis, we need to obtain an algorithm with the ability to identify local outliers, namely the LOF algorithm. In a given data set, if any data point has dense points in its local range, then this data point is considered to be a normal data point. Otherwise, it may be an outlier.
[0102] Suppose there is a data set sample C, assuming there are n samples in total, the data dimension is m, for
[0103]
[0104] The Euclidean distance between any two points can be expressed as:
[0105]
[0106] The kth closest point P is the kth closest distance point of O, that is, d k (O) is the kth distance of point O, which is defined as follows:
[0107] d k (O) = d(O,P)
[0108] If the following conditions are met:
[0109] (1) There are at least k points P′∈C\{O} in the set such that d(O,P′)≤d(O,P); where C\
[0110] {O} represents the difference set of C with respect to {O}, and \ is the difference set symbol;
[0111] (2) There are at most k-1 points P′∈C\{O} in the set such that d(O,P′) <d(O,P)。
[0112] At this time, point P is the kth point closest to O.
[0113] For example, Figure 7 As shown, C is a set of points centered at O. 6 or P 7 They are all the closest 6th point. There are 5 points inside the 6th distance, and there are 2 points at the 6th distance.
[0114] RD(P,O) is a custom function that obtains the maximum distance. The distance of point P with respect to data point O can be calculated according to equations (3)-(4). For example, Figure 5 In the example, we set the parameter k=6, P 5 The reachable distance is d k=6 (O), P 10 The reachable distance is dist(P 10 ,O), where P 10 is outside the k distance, in which case the Euclidean distance dist(P 10 ,O). Therefore, the reachable distance of point P relative to point O is the maximum value of the k distance of point O and the Euclidean distance between P and O.
[0115] reach-dist k (P,O)=max(d k (O),dist(P,O))(3)
[0116] Right now:
[0117] RD(P,O)=max(d k (O),dist(P,O))(4)
[0118] Here, dist(·) represents the Euclidean distance.
[0119] Local reachability density (LRD) is the inverse of the average reachable distance from P to its neighboring points. Intuitively, according to the LRD formula, the larger the average reachable distance (that is, the farther the neighboring points are from point P), the smaller the density of neighboring points around the point. Therefore, LRD expresses the distance from a point to the nearest point group. The lower the LRD value, the farther the nearest point group is from the point.
[0120]
[0121] Among them, N d (P) represents the set of neighboring points of P, |N k (P)| represents the capacity of the set.
[0122] Local outlier factor (LOF) is the ratio of the average LRD of P's k neighbors to P's LRD:
[0123]
[0124] Furthermore, if point P is not an outlier, then the ratio of the average LRD of the neighboring points is approximately equal to the LRD of point P, because the density of a point is roughly equal to the density of its neighboring points. In this case, LOF is almost equal to 1. On the other hand, if a point is an outlier, the LRD of the point is less than the average LRD of the surrounding neighboring points. Then the LOF value will be very high. Usually, if LOF>1, it is considered to be an outlier, but this is not always correct. Assuming that we know that there is only one outlier in the data, then the point corresponding to the maximum value of all LOF values will be considered an outlier. Obviously, the LOF value of a point directly expresses the degree of abnormality of the point.
[0125] Improved LOF algorithm——Attention Enhanced AE-LOF algorithm:
[0126] "Focus" is used as the core word to explain the AE-LOF algorithm. For example, on a social network, if A follows many people {B, C, D, E, F, ...}, it cannot be considered that A's attention is high if only A follows many people. Only when many people {B, C, D, E, F, ...} also follow A, can it be considered that A's attention is high.
[0127] Therefore, the AE-LOF algorithm can be explained as follows: Assume that there is a data set C consisting of several data points, and the data points of the data set can be correlated with each other. Data set C = {p 1 , p 2 ,…,p n}, C contains n data points. Data point p i The radius of attention is R i Data point p i and p j The Euclidean distance between them is d(p i , p j ).
[0128] The set of associated points is A i , for each data point p i , define the set of associated points A i as follows:
[0129] A i ={p j ∈C\{p i}|d(p i , p j )≤R i} (7)
[0130] This set contains all the i The radius R i Other data points within .
[0131] Frequency statistics: For each point p in the data set j , initialize a frequency counter F j = 0. For each data point p i , check every other point p j Is it in A i If p j In A i In, F j Increase by 1. For each i and each j ≠ i, if p j ∈A i , then F j =F j +1. After traversing all the points, each F j
[0132] will represent the point p j As the total number of times the associated point appears within the radius of all other points. In summary, for each point p j The number of occurrences of global association points F j This can be calculated by iterating over the dataset and updating the counter:
[0133]
[0134] Among them, 1 is the indicator function, when p j In A i The value is 1 when it is in, otherwise it is 0.
[0135] In order to adjust the LOF value of each point to a new value that takes into account its "attention", F is integrated on the basis of the original LOF algorithm. j Therefore, the present invention is implemented by adjusting the LRD calculation method of each point. Specifically, assuming F j The smaller the value, the more points p j The more likely it is to be isolated, the higher its abnormality should be. Then, by introducing F j The method of using the value to improve the LOF value of these points in disguised form, thereby improving the abnormality of these points.
[0136] Adjusted AE-LOF(p j ) is defined as:
[0137]
[0138] Wherein, Z is the internal adjustment coefficient, which is usually a positive integer, and is generally 1. w is the external adjustment coefficient, which is usually a positive integer, and can generally be w=Z×k.
[0139] When w = Z × k, there are two main calculation results: (1) When point p j If it is not an outlier, then F j is equal to or close to k, then can be simplified to at this time The calculated value of is approximately equal to 1. Therefore, in this case, AE-LOF(p j ) is similar to the original LOF(p j ) is almost unchanged, ensuring the original LOF(p j )’s results. (2) When point p j When it is an outlier, then F j will be less than k, then can be simplified to But at this time The calculated value of will become much larger. Therefore, in this case, AE-LOF(p j ) will be greater than the original LOF(p j ) calculation results, ensuring AE-LOF(p j ) can effectively filter out outliers.
[0140] After the above steps, click p jThe local density of is adjusted by the number of occurrences of its global associated points, so that those points that are less concerned are promoted in the abnormality evaluation. This not only strengthens the abnormality analysis of each point, but also considers the new value of "attention degree" F j (the number of occurrences of the associated point) and Z, w (internal and external adjustment coefficients) provide more flexibility and adaptability to handle different data and scenarios.
[0141] (A1), original LOF algorithm results
[0142] For mixed signal data, the LOF algorithm can be used to obtain the LOF value of each point to be detected. By setting the outlier detection threshold, the outliers can be screened out. Generally speaking, after obtaining the outliers, it is also necessary to evaluate the degree of abnormality of each outlier, which is conducive to visual inspection, display and subsequent evaluation of related data. It should be emphasized that the evaluation of the degree of abnormality of the points to be tested is mainly completed by comparing the LOF value, so the LOF value is a key parameter. The conventional outlier abnormality evaluation based on the LOF algorithm has defects. Figure 8 The data in is used as a typical object to illustrate the defects. Figure 8 The figure shows the outlier anomaly score after calculation based on the LOF algorithm. The outliers are mainly divided into three areas, among which No.1 in area C has the smallest LOF value. In fact, compared with the outlier set {No.75, No.40, No.80, No.57, No.60, No.62}, outlier No.1 is about 4000 points away from the center line, and the degree of anomaly of outlier No.1 is obviously more obvious. Therefore, the evaluation of outlier anomaly based on the LOF algorithm has defects.
[0143] First, take the outlier No.40 in area A as an example to analyze the reason why the LOF value of this point is large. The LOF algorithm is mainly a density-based outlier detection algorithm. To evaluate the degree of abnormality of any data point in the test data set, it is usually necessary to refer to the degree of abnormality of the k neighboring points of the data point. For the outlier No.40 located in area A, its k neighboring points (k=10) mainly include three parts, namely the outliers {No.75, No.80} in area A, the outliers {No.62, No.60, No.57} in area B, and the data points {No.81, No.70, No.74, No.69, No.73} located in the center yellow area (Center Area) (because Figure 8 and Fig. 9 The display space is limited, so this part is not included in Figure 8 and Fig. 9The calculation results show that the LRD values of the data points {No.81, No.70, No.74, No.69, No.73} in the center yellow area are all on the order of 1×10 -2 , while the LRD magnitudes of outliers {No.75,No.80} and {No.62,No.60,No.57} are both 1×10 -4 Obviously, due to the large difference in the order of magnitude of the above data points, the LRD mean of the k neighboring points (k=10) of the outlier point No. 40 in area A is significantly affected by the five data points located in the center yellow area (Center Area), resulting in a higher LOF calculated value for the outlier point No. 40 in area A.
[0144] Secondly, we analyze the reason why the LOF value of outlier No.1 in area C is small. For outlier No.1 in area C, its k neighbor points (k=10) are all distributed in the central yellow area and concentrated in area C-1. As can be seen from area C-1 in the figure, the data points in area C-1 are far away from the surrounding points, and their LRD values are smaller than other data points in the center yellow area (Center Area), which will lead to a lower LOF calculation value for outlier No.1 in area C.
[0145] Through the above analysis, the LOF value of outlier No.1 in area C is smaller than the LOF value of outlier No.40 in area A. Similarly, the LOF value of the outlier set {No.75, No.80, No.57, No.60, No.62} in areas A and B is larger than the LOF value of outlier No.1 in area C.
[0146] Table 1 Schematic diagram of the calculation process of LOF values of each outlier in the main area
[0147]
[0148] (A2) AE-LOF algorithm results
[0149] For mixed signal data, Table 2 shows the correlation frequency statistics of outliers in regions A, B, and C. Sorted by the statistical value of correlation frequency: region C < region B < region A. Based on the statistical value of correlation frequency, the LOF value of the outliers is recalculated by the AE-LOF algorithm and plotted, as shown in the figure. Fig. 9 As shown. Fig. 9 It can be seen that the LOF value of outlier No. 1 in area C increases significantly compared to the LOF value of outlier in area A, and the degree of abnormality increases significantly. At the same time, the LOF value of outlier in area B also decreases significantly compared to the LOF value of outlier in area A.
[0150] Table 2
[0151]
[0152] from Fig.10 From the locally enlarged detail diagram in , it can be seen that for non-outliers, the calculated values of the LOF algorithm and the AE-LOF algorithm are approximately equal, and the overall change trend is the same. For outlier No.1, the calculated value based on the AE-LOF algorithm is 80 times higher than that of the LOF algorithm. For outliers {No.62, No.60, No.57}, it is increased by about 3 times. For outliers {No.40, No.80, No.75}, it is increased by about 7 times. The detection value range of the AE-LOF algorithm is wider, and it can capture more significant outliers (such as 1620.54 corresponding to outlier No.1 in Table 2), which shows that the AE-LOF algorithm is more sensitive in detecting anomalies, and the AE-LOF algorithm can identify more obvious outliers in anomaly detection.
[0153] In the outlier detection based on the LOF algorithm, the LOF result of any point P is affected by the neighboring points of the point. When some of the neighboring points of point P are slightly abnormal, from a global perspective, the LOF value of P will be smaller to a certain extent, which may cause some outlier anomalies to be ignored or covered up. The local outlier detection algorithm based on enhanced attention density (AE-LOF algorithm) can make up for this problem of the LOF algorithm to a certain extent when the neighboring points are abnormal. Due to the confidentiality of aerospace detection, experimental data cannot be obtained and disclosed in large quantities, and also contains many complex abnormal situations in redundant object detection. The above multiple reasons have led to the current recognition ability of the mixed signal of redundant object detection to be further improved. In the later stage, it is planned to increase the number of field experiments, obtain more experimental data, and further study the outlier value detection of the redundant object detection signal. The experimental results show that, in general, for outliers, the AE-LOF calculation value is several times higher than the LOF calculation value. AE-LOF is more sensitive in detecting anomalies, has a wider range of detection values, and can capture more significant anomalies. The AE-LOF algorithm is effective in detecting and identifying the abnormality of some outliers in the aerospace relay redundant object detection signal.
[0154] In fact, the present invention may also adopt any other improved LOF algorithm to implement evaluation.
[0155] (B) Clustering based on k-CAD is performed for the peak point coordinates of the pulses corresponding to all segmented segments on the xy coordinate axis:
[0156] The data point set corresponding to the peak points of the pulses on all the segmented segments on the xy coordinate axis is recorded as data set C, where the data point is xi , assuming there are n samples in total, and the data dimension of each sample (actually the two-dimensional coordinates of the point) is m:
[0157]
[0158] The goal of k-means clustering is to divide the data set C into k categories (clusters), and the center point of each category is denoted by μ j (j=1,2,…,k). The objective function of k-means clustering is to minimize the sum of squared errors (Sum of SquaresError, SSE) within the cluster:
[0159]
[0160] Among them, δ ij is the indicator function, when point x i When it belongs to category j, δ ij =1, otherwise δ ij =0. By calculating the SSE under different k values and drawing a relationship graph between SSE and k, the "elbow" position (that is, the inflection point position) is selected as the optimal number of clusters. The specific steps are: (1) Select a k value range. After research and verification by the present invention, for the aerospace sealed relay excess detection signal, k generally does not exceed 5; (2) Perform k-means clustering on each k value and calculate the corresponding SSE; (3) Draw a relationship graph between k value and SSE; (4) Observe the "elbow" position and the number of data points contained in each cluster to determine the optimal k value.
[0161] Based on the optimal k value, the k-means algorithm is used to cluster the data set C and obtain the category label L of each point = [l 1 ,l 2 ,…,l n ], where l i Represents the signal point x i Category, l i ∈{1,2,…,k}. For each category j (j=1,2,…,k), extract all sample points belonging to this category and record them as set C j The center μ of category j j The calculation formula is:
[0162]
[0163] Among them, |C j | is the number of sample points of category j.
[0164] The calculation formula of category j based on the k-CAD algorithm is:
[0165]
[0166] According to long-term practical test experience, if for all j, the condition k-CAD(C j )<1000, the clustering results of the redundant detection signals are considered meaningful.
[0167] For a component signal data, such as Fig.11 As shown in (a), before the multi-component signal data points are clustered, the calculated value of the original center point average distance is usually more than 1000. In this case, the calculated value of the original center point average distance cannot accurately reflect the cluster characteristics of the multi-component signal data, which may lead to misclassification. To solve this problem, the present invention combines the k-means algorithm and the center average distance (CAD) algorithm to propose a k-cluster center average distance method (k-means Center Average Distance, k-CAD). The algorithm can effectively measure the volatility of data points in the signal, thereby evaluating the complexity and composition of the signal.
[0168] Compared with the simple standard deviation algorithm (Standard Deviation, STD), the center average distance algorithm (Center Average Distance, CAD) is applicable to various data types and distributions, especially for non-normally distributed data. In addition, the CAD algorithm is insensitive to outliers and will not be seriously affected by extreme values. Fig.11 (b) It can be seen that the calculation result based on the k-means Center Average Distance (k-CAD) method is 64.5, which is Fig.11 The calculation result in (a) is significantly lower than that in (a) by a factor of about 40. Therefore, it is effective to determine the number of clusters in a multi-component signal based on the k-CAD method.
[0169] The use of the k-means algorithm requires the number of clusters to be determined in advance. The present invention uses the elbow method to determine the optimal number of clusters. Although in many practical applications, the choice of "elbow" is very vague because it does not contain an elbow with a clear inflection point. However, in the redundant object detection signal, the attempt to capture the curvature using the "slope" is very clear because the preprocessed component signals usually show obvious cluster characteristics. In addition, in redundant object detection, when the number of data points in a cluster is less than 15, we consider the data cluster to be a redundant signal because the number of component signals is generally higher than 15. Subsequently, the k-CAD algorithm will skip this clustering to reduce calculation time and improve calculation efficiency. Fig.12 As shown, Fig.11Taking the data in as an example, when the number of clusters is 2, the maximum deviation multiple is close to 40, which is an obvious inflection point - "elbow". As the number of clusters increases, the maximum deviation multiple tends to be stable and is significantly lower than 40. It should be further explained that the maximum deviation multiple is a parameter used to calculate the ratio of the differences between adjacent elements in a vector or matrix, which is used to measure the difference between two variables. In the present invention, based on the experience of redundant object detection, if the maximum deviation value multiple between adjacent elements exceeds 10, it is considered to be a possible obvious inflection point - "elbow", and the number of clusters at this time is usually the optimal number of clusters.
[0170] For all points on the xy coordinate axis, the specific process of measuring signal classification is as follows: Fig.13 As shown, combined with long-term experience in the identification of redundant object detection signals, the present invention proposes a new closed-loop classification system for analyzing and identifying the types of redundant object detection signals. The classification system takes the preprocessed signal as input, combines multiple algorithms and multi-level rules, and realizes accurate identification of the types of redundant object detection signals. The main functions of the classification system include: 1) Input check of the preprocessed signal, which can make the input data meet the processing requirements. 2) Calculate the pulse duty cycle. This process can quickly screen out excessive component signals, reduce redundant analysis, and improve identification efficiency. 3) The AE-LOF algorithm is combined with the initial CAD algorithm to deeply analyze the composition of the signal, thereby accurately identifying single-component signals and mixed signals I. 4) The AE-LOF algorithm is combined with the k-CAD algorithm to accurately identify redundant particle signals, mixed signals II and multi-component signals. It should be reiterated that the k-CAD algorithm determines the optimal number of clusters through the elbow method, and then calculates the CAD of each cluster. k (xi) value to evaluate the volatility and composition complexity of the detection signal. It should be noted that in the process of finding the optimal number of clusters, the minimum number of data points of a cluster is set to no less than 15, otherwise the cluster will not be identified as a qualified cluster. According to the experience of redundant object detection, the number of pulses of the general component signal is not less than 15 within a vibration period of 5s. Setting this screening rule can quickly and effectively find the optimal k value. 5) For some rare and complex signals, the classification system will prompt that manual identification is required to form a closed loop of the identification process to ensure the reliability of the final result. In summary, this classification system provides an efficient and accurate redundant object detection signal identification tool through the combination of multiple algorithms and multi-level rules.
[0171] The specific implementation steps are as follows:
[0172] (1) Identification 1: Calculate the pulse duty cycle and perform preliminary identification:
[0173] Step 1: For the data points in step 2, ensure the validity of the input data;
[0174] Step 2. Calculate the pulse duty cycle D (Duty cycle). When D ≥ 65%, it is an oversized component signal; when D < 65%, continue to identify it. It should be noted that the oversized component signal is a special signal, usually generated by a component that vibrates violently. Because this type of signal will cover or interfere with the information of other types of signals, it is impossible to effectively identify the components of the signal. Therefore, when D ≥ 65%, there is a high possibility that the signal is an oversized component signal.
[0175] (2) Identification 2: The result of the CAD algorithm is recorded as CAD(xi) (it can be considered that k is first set to 1 in the k-CAD algorithm, and the k-CAD algorithm corresponding to the optimal k value is used in the subsequent steps). The identification is performed based on the results of CAD(xi) and the AE-LOF algorithm:
[0176] Step 3, when CAD(xi)≤1000, and the AE-LOF algorithm detection result has no abnormal value, the identification result is: single component signal;
[0177] Step 4. When CAD(xi)≤1000 and the AE-LOF algorithm detection result has an abnormal value, the identification result is: the single component signal is mixed with the redundant particle signal, that is, mixed signal I.
[0178] (3) Judgment 3: The result calculated by the k-CAD algorithm is recorded as CAD k (xi), based on CAD k (xi) and the results of the AE-LOF algorithm are used to judge:
[0179] Step 5. If CAD k (xi)≤1000, and the AE-LOF algorithm detection result has no abnormal value, and the identification result is: multi-component signal;
[0180] Step 6. If CAD k (xi)≤1000, and the AE-LOF algorithm detection result has an abnormal value, and the identification result is: a mixture of multi-component signals and redundant particle signals, that is, mixed signal II;
[0181] Step 7, if CAD k (xi)>1000, and the AE-LOF algorithm detection result has an abnormal value, and the identification result is: redundant particle signal;
[0182] Step 8. If CAD k (xi)>1000, and there are no abnormal values in the detection results of the AE-LOF algorithm. The identification result is: manual identification is required.
[0183] In order to better illustrate the classification rules and logic of the classification system of the present invention, five typical types of unwanted object detection signals are analyzed and displayed one by one. Specifically, the five typical signals mainly include single-component signals, mixed signal type I, unwanted object particle signals, multi-component signals, and mixed signal type II. The calculation results obtained based on the AE-LOF algorithm and the k-CAD algorithm are directly marked in the schematic diagram for easy comparison and analysis. In addition, there are different specific problems in the identification of different signal types. In order to better illustrate these problems, we have made different annotations and explanations on different schematic diagrams. In summary, it can better help readers understand the classification rules, algorithm innovations, and the original intention of the early data preprocessing of the classification system in the present invention.
[0184] Fig.14 The scatter plots of single component signals at two different horizontal axis scales are shown. Fig.14 In (a), the range of the horizontal axis is expanded for two main purposes: (1) to match the interface of the PIND instrument detection. (2) to match the number of complete sampling of the signal acquisition system. In addition, in the manual identification stage, Fig.14 The horizontal axis range of (a) is also the range that needs to be used for dynamic display of detection results during manual identification. Fig.14 In (b), the range of the horizontal axis is reduced, the main purpose of which is to clearly show the location of the center lines of different clustering areas. Fig.14 (a) It can be seen that when all data points are clustered according to k = 1 and 2, the corresponding values based on the k-CAD algorithm are 95.40 and 74.14, respectively, and both are lower than 1000. From the comparison results, it can be seen that the changes in the k-CAD values corresponding to different cluster numbers k are not obvious, and there are no outliers. In summary, the signal can be identified as a single component signal.
[0185] Fig.15 Shown is a scatter plot of the excess particle signal. Fig.15 (a) It can be seen that since the initial CAD value is 4035.60, which exceeds the threshold of 1000 in judgment 2, it is necessary to use the k-CAD algorithm to calculate and analyze again. In addition, from the partially enlarged image, it can be seen that some data points are difficult to obtain through manual observation due to image superposition. In fact, the same signal image coverage problem also exists when manually judging the detection waveform on the PIND instrument. This once again reflects the necessity and scientific nature of using the judgment algorithm. Fig.15 (b) It can be seen that when k = 2, the k-CAD value is 1844.10, which exceeds the threshold of 1000 in judgment 3, and there are multiple outliers. In summary, the signal can be identified as a redundant particle signal. Fig.15 (b) It can be seen that the outlier point obtained by the AE-LOF algorithm is P3 -P 5 However, from the position of each data point and their distances to each other, P 1 and P 2 points are more likely to become outliers. In order to fully analyze this problem, we calculated P 1 -P 5 The LOF value and AE-LOF value of the point are calculated and the corresponding value is marked on the point. Through comparative analysis of the calculated values, the main reasons are as follows: (1) P 1 Point and P 2 The AE-LOF value of a point has a strong dependence on the surrounding data points; (2) P 1 and P 2 The distribution of data points around the point is loose; (3) P 1 and P 2 The points around the point are of high interest. (4) The order of magnitude of the ordinate of each point is much lower than that of the abscissa, which is also an important reason for the difference between the visual observation results and the calculation results based on the algorithm. Generally speaking, the larger the outlier value of a point, the farther it is from the center line. However, the evaluation of actual outliers often needs to consider multiple factors such as the distribution, number and attention of the surrounding points.
[0186] The initial CAD value is 678.74, which is lower than the threshold of 1000 in the second judgment. Fig.16 (b) It can be seen that multiple outliers are captured by the AE-LOF algorithm, among which P 1 The outlier value of the point is the largest, the calculated attention value is the lowest, and the degree of anomaly is the highest. This shows that the AE-LOF algorithm has obvious superiority in detecting outliers located at the edge of the cluster. At the same time, when k=2, the k-CAD value is 355.44, which is lower than the threshold of 1000 in judgment 3. Obviously, continuing the calculation of the k-CAD value is redundant, which once again reflects the scientific nature of using the initial CAD value as a priority judgment condition in the classification system of the present invention. In summary, the signal can be identified as a mixture of a single component signal and a redundant particle signal - mixed signal I.
[0187] Fig.17 The displayed image is a mixed scatter plot of multi-component signals and excess particle signals. Fig.17 (a) It can be seen that the initial CAD value is 1168.80, which is higher than the threshold of 1000 in judgment 2, and further calculation using the k-CAD algorithm is required. Figure 4(b) It can be seen that when k = 2, the k-CAD value is 522.25, which is lower than the threshold of 1000 in judgment 3, and multiple outliers can be obtained through the AE-LOF algorithm. In summary, the signal can be identified as a mixture of multi-component signals and redundant particle signals - mixed signal II. It should be noted that the optimal clustering result of multi-component signals may require multiple iterative calculations to obtain, because different k values will obtain different k-CAD values. If there are several outliers with similar positions, but they are far away from the main data cluster of the component signal as a whole, the k-mean algorithm may mistakenly regard these outliers as a data cluster, and the k-CAD value may still be greater than 1000. Therefore, in the iterative calculation process, we set that when the number of data points in a cluster is less than 15, it will not be identified as a qualified cluster. According to the process of the classification system, this signal type may be identified as requiring manual identification. This situation is special, and the identification result is acceptable.
[0188] Fig.18 Shown is a scatter plot of a multi-component signal. Fig.18 (a) It can be seen that at that time, the initial CAD value was 1303.20, which was higher than the threshold of 1000 in the judgment 2. Fig.18 (b) It can be seen that when k = 2, the k-CAD value is 53.25. Obviously, for multi-component signal types, the k-CAD values corresponding to different k values vary significantly, and this variation is often more than 10 times. Generally speaking, the vibration frequency of the component signal is relatively stable, and occasionally there is a shift in the frequency center point, but the overall absolute value of the deviation is less than 1000. Otherwise, if the vibration of the component is too severe and the displacement is too flexible, it will be more harmful to the quality of the product and will not meet the product reliability design requirements.
[0189] In summary, we have explained the identification process of the five signal categories one by one, which helps readers understand the classification rules and design logic of the classification system. It is worth noting that the systematic strategy of the present invention is also valuable for the identification of non-periodic signals and periodic signals containing abnormal signals.
[0190] In order to verify the accuracy of the classification system of the present invention, we selected 800 groups of signal sample data from the database of the research group according to the random sampling rule as the identification object. These signal sample data are as follows: (1) 200 groups of known single component signal samples; (2) 200 groups of known mixed signal I samples; (3) 200 groups of known redundant signal samples; (4) 100 groups of known mixed signal II samples; (5) 50 groups of multi-component signal samples; (6) 50 groups of known oversized component signal samples. The identification statistical results of these signal data are shown in Table 3.
[0191] Table 3 Identification results and comparative statistics based on the classification system
[0192]
[0193] Note: (1) * :The number of signal samples of mixed signal II is small, because the actual probability of occurrence is also small. (2)**: The number of samples of multi-component signal is small, because the actual probability of occurrence is also small. (3)***: The pulse duty cycle threshold D used in this test is 65%, which comes from long-term data statistics. (4)△: The main parameters of data statistics are accuracy and improvement ratio. Since random rules are used as the selection method for test sample data, the differences and contingencies brought by different signal categories and different numbers of test samples are not considered for the time being. (5)△△: Since the old algorithm cannot reclassify the component signal type, the overall recognition accuracy of the component signal is used instead.
[0194] As can be seen from the table, 193 groups of single-component signals were correctly identified with an accuracy of 96.50%; 190 groups of mixed signals I were correctly identified with an accuracy of 95.00%; 192 groups of redundant signals were correctly identified with an accuracy of 96.00%; 86 groups of mixed signals II were correctly identified with an accuracy of 93.00%. 46 groups of multi-component signals were correctly identified with an accuracy of 92.00%; 49 groups of oversized component signals were correctly identified with an accuracy of 98%.
[0195] It should be additionally explained that this recognition accuracy is the result of signal confusion recognition with 6 types of classification objects. Other previous studies are not based on the results of 6 types of signal confusion recognition. When the prior art is based on 6 types of signal confusion recognition, its disclosed accuracy will inevitably be greatly affected. The present invention obtains the above recognition results in 6 types of signal confusion recognition, which is sufficient to illustrate the effect of the present invention.
[0196] In order to reflect the continuity and improvement effect in the classification system research, a comparison is made with CN117009858B (the previous research method of the inventor's team). It is worth noting that the average accuracy of the strategy proposed in the present invention is 95.08%, while the average accuracy of the old method is only 89.10%, and the average accuracy is improved by about 6.71%. In addition, in order to more clearly show the comparison results between the two methods, we plot the strategy of the present invention and the previous research results into line graphs according to the signal category. Fig.19(On the horizontal axis in the figure, 1 is a single-component signal, 2 is a mixed signal I, 3 is a redundant signal, 4 is a mixed signal II, 5 is a multi-component signal, and 6 is an oversized component signal) It can be seen that compared with the previous method, the strategy of the present invention has improved the recognition accuracy on the six types of signal sample data, among which the recognition accuracy for mixed signal I has increased the most, reaching 13.42%. Specific implementation method 2:
[0198] This embodiment is a classification system for aerospace sealed relay excess detection signal, which is actually a classification system composed of program modules corresponding to a classification method for aerospace sealed relay excess detection signal, including:
[0199] Pulse extraction module: extract pulses from PIND signals collected during redundant object detection of aerospace relays;
[0200] Pulse segmentation module: divide the extracted pulse signal into frames;
[0201] Signal folding and coordinate projection module: fold the framed signal; then use the coordinates of the peak point of each pulse in the frame where the pulse is located to represent the pulse, and display the peak points of the pulses on all frame segments on the xy coordinate axis, where the x-axis is the length direction of each frame signal and the y-axis is the amplitude direction of each frame signal;
[0202] Classification module of detection signal: Based on the peak point coordinates of the pulses corresponding to all the frame segments on the xy coordinate axis, multiple data points in the xy coordinate system are obtained. For multiple data points, k-CAD is used for clustering. At the same time, the abnormality of outliers is evaluated based on the peak point coordinates of the pulses corresponding to all the frame segments on the xy coordinate axis. Then, based on the clustering results and the abnormality, the classification system is used to classify the detection signal.
[0203] The classification module of the detection signal includes a clustering result calculation unit, an outlier abnormality evaluation unit, and a classification unit; wherein,
[0204] Clustering result calculation unit: based on the peak point coordinates of the pulses corresponding to all the frame segments on the xy coordinate axis, multiple data points in the xy coordinate system are obtained, and the CAD values of all the points are calculated for the multiple data points. Meanwhile, for the multiple data points, based on the optimal clustering number k, k-CAD is used to perform clustering and calculate the k-CAD value;
[0205] Outlier abnormality degree assessment unit: assesses the abnormality degree of the outlier based on the peak point coordinates of the corresponding pulses on all frame segments on the xy coordinate axis;
[0206] Classification and identification unit: classifies the detection signal according to the result of the clustering result calculation unit and the result of the outlier abnormality evaluation unit. Specific implementation method three:
[0208] This embodiment is a method for clustering redundant object detection signals for aerospace sealed relay redundant object detection signals, comprising the following steps:
[0209] Pulse extraction is performed on the PIND signal collected in the detection of redundant objects in aerospace relays, and the extracted pulse signal is framed; then each framed signal is used as a data point sample x i , convert each framed signal into an m-dimensional vector [x i1 ,x i2 ,…,x im ], use [x i1 ,x i2 ,…,x im ] as the data point x i The coordinates of
[0210] In this embodiment, each framed signal is taken as a data point sample x. i , convert each framed signal into an m-dimensional vector [x i1 ,x i2 ,…,x im ], the peak points of the pulses on all the frame segments are displayed on the xy coordinate axis, the x axis is the length direction of each frame signal, and the y axis is the amplitude direction of each frame signal; based on the peak point coordinates of the pulses corresponding to all the frame segments on the xy coordinate axis, multiple data points in the xy coordinate system are obtained,
[0211] The set of data points corresponding to the peak points of the pulses corresponding to all segmented segments on the xy coordinate axis is recorded as data set C:
[0212] C={x i |i=1,2,…,n},x i =[x i1 ,x i2 ,…,x im ]
[0213] Among them, x i represents a data point sample, [x i1 ,x i2 ,…,x im ] is x i The coordinates of , n is the number of data point samples;
[0214] Based on the optimal number of clusters k, the k-means clustering algorithm is used to cluster the data set C to obtain the category label L of each point = [l1 ,l 2 ,…,l n ], where l i Represents the signal point x i Category, l i ∈{1,2,…,k}; for each category j, j=1,2,…,k, extract all sample points belonging to this category, recorded as set C j ; The center μ of category j j The calculation formula is:
[0215]
[0216] Among them, |C j | is the number of sample points of category j;
[0217] The k-CAD clustering result of category j is
[0218] k-CAD(C j ) is used for the classification of redundant object detection signals of aerospace sealed relays.
[0219] The above calculation examples of the present invention are only used to explain the calculation model and calculation process of the present invention in detail, and are not intended to limit the implementation methods of the present invention. For ordinary technicians in the relevant field, other different forms of changes or modifications can be made based on the above description. It is impossible to list all the implementation methods here. All obvious changes or modifications derived from the technical solution of the present invention are still within the scope of protection of the present invention.
Claims
1. A method for classifying redundant object detection signals of aerospace sealed relays, characterized in that: The following steps are involved: Pulse extraction is performed on the PIND signal collected in the detection of redundant objects in aerospace relays, and the extracted pulse signal is framed; then the framed signal is folded; The pulse is represented by the coordinates of the peak point of each pulse in the sub-frame where the pulse is located, and the peak points of the pulses on all sub-frame segments are displayed on the xy coordinate axis, where the x-axis is the length direction of each frame signal and the y-axis is the amplitude direction of each frame signal; Based on the peak point coordinates of the pulses corresponding to all the frame segments on the xy coordinate axis, multiple data points in the xy coordinate system are obtained. For the multiple data points, k-CAD is used for clustering. At the same time, the abnormality of the outliers is evaluated based on the peak point coordinates of the pulses corresponding to all the frame segments on the xy coordinate axis. Then, the detection signal is classified based on the clustering results and the abnormality.
2. The method for classifying redundant object detection signals of aerospace sealed relays according to claim 1, characterized in that: In the process of framing the extracted pulse signal, the length of the vibration cycle of the vibration table is used as the length of the framing window for framing.
3. The method for classifying redundant object detection signals of aerospace sealed relays according to claim 1, characterized in that: Before the pulse signal is framed, adjacent pulses whose time interval is less than a set threshold are merged into one pulse.
4. A method for classifying aerospace sealed relay redundant object detection signals according to any one of claims 1 to 3, characterized in that: For multiple data points in the xy coordinate system, the process of clustering using k-CAD includes the following steps: The set of data points corresponding to the peak points of the pulses corresponding to all segmented segments on the xy coordinate axis is recorded as data set C: C={x i ∣i=1,2,…,n},x i =[x i1 ,x i2 ,…,x in ] Among them, x i represents a data point sample, [x i1 ,x i2 ,…,x im ] is x i The coordinates of , n is the number of data point samples; Based on the optimal number of clusters k, the k-means clustering algorithm is used to cluster the data set C to obtain the category label L = [l1, l2, ..., l n ], where l i Represents the signal point x i Category, l i ∈{1,2,…,k}; for each category j, j=1,2,…,k, extract all sample points belonging to this category, recorded as set C j ; The center μ of category j j The calculation formula is: Among them, |C j | is the number of sample points of category j; The k-CAD clustering result of category j is 5. The method for classifying aerospace sealed relay redundant object detection signals according to claim 4, characterized in that: The optimal number of clusters k is determined according to the following steps: The k-means clustering algorithm is used to divide the data set C into k categories. The SSE under different k values is calculated, and the relationship between the k value and the sum of squared errors within the cluster is plotted; the k value at the inflection point is used as the final determination of the optimal number of clusters.
6. The method for classifying aerospace sealed relay redundant object detection signals according to claim 5, characterized in that: The AE-LOF algorithm is used to evaluate the abnormality of outliers based on the peak point coordinates of the pulses corresponding to all frame segments on the xy coordinate axis. The specific process includes: For the peak point coordinates of the pulses corresponding to all the frame segments on the xy coordinate axis, the associated point set A is first obtained. i ; For each data point p i , the set of associated points A i as follows: A i ={p j ∈C\{p i }∣d(p i ,p j )≤R i } This set contains all the i The radius R i Other data points within d k (·) represents the k-neighbor distance; For each point p in the data set j , initialize a frequency counter F j =0; for each data point p i , check every other point p j Is it in A i If in then F j =F j +1; After traversing all points, each F j will represent the point p j The total number of times the associated point appears within the radius of all other points; Then calculate AE-LOF(p j ): Among them, N k (·) represents the set of neighbor points, |N k (P)| represents the set capacity; LRD(·) represents the inverse of the average reachable distance from a point to its surrounding neighboring points; Z is the internal adjustment coefficient, and w=Z×k is the external adjustment coefficient; Using AE-LOF(p j ) evaluates the abnormality of outliers based on the peak points of the corresponding pulses on all frame segments on the xy coordinate axis.
7. The method for classifying aerospace sealed relay redundant object detection signals according to claim 6, characterized in that: The process of classifying the detection signal based on the clustering results and the evaluation of the degree of abnormality includes the following steps: For multiple data points in the xy coordinate system, calculate the pulse duty cycle D of the pulse corresponding to each point. When D ≥ the duty cycle threshold, it is identified as an excessive component signal; otherwise, continue to identify; Then calculate all the points Among them, |C| represents the capacity of the data set C corresponding to all points, μ j Represents the C mean of the data set; when CAD(xi)≤clustering threshold, and the AE-LOF algorithm detection result has no outliers, the identification result is a single-component signal; when CAD(xi)≤clustering threshold, and the AE-LOF algorithm detection result has outliers, the identification result is a mixture of a single-component signal and a redundant particle signal; otherwise, continue to identify; The calculation result of the k-CAD algorithm corresponding to the optimal number of clusters k is recorded as CAD k (xi) If CAD k (xi)≤clustering threshold, and the AE-LOF algorithm detection result has no outliers, and the identification result is a multi-component signal; if CAD k (xi)≤clustering threshold, and the AE-LOF algorithm detection result has an outlier, and the identification result is a mixture of multi-component signals and redundant particle signals; if CAD k (xi)>clustering threshold, and the AE-LOF algorithm detection result has an abnormal value, and the identification result is an extraneous particle signal.
8. An aerospace sealed relay redundant object detection signal classification system, characterized in that: include: Pulse extraction module: extract pulses from PIND signals collected during redundant object detection of aerospace relays; Pulse segmentation module: divide the extracted pulse signal into frames; Signal folding and coordinate projection module: fold the framed signal; then use the coordinates of the peak point of each pulse in the frame where the pulse is located to represent the pulse, and display the peak points of the pulses on all frame segments on the xy coordinate axis, where the x-axis is the length direction of each frame signal and the y-axis is the amplitude direction of each frame signal; Classification module of detection signal: Based on the peak point coordinates of the pulses corresponding to all the frame segments on the xy coordinate axis, multiple data points in the xy coordinate system are obtained. For multiple data points, k-CAD is used to perform clustering. At the same time, the abnormality of outliers is evaluated based on the peak point coordinates of the pulses corresponding to all the frame segments on the xy coordinate axis. Then, the detection signal is classified based on the clustering results and the abnormality.
9. The aerospace sealed relay redundant object detection signal classification system according to claim 8, characterized in that: The classification module of the detection signal includes a clustering result calculation unit, an outlier abnormality evaluation unit, and a classification unit; wherein, Clustering result calculation unit: based on the peak point coordinates of the pulses corresponding to all the frame segments on the xy coordinate axis, multiple data points in the xy coordinate system are obtained, the CAD values of all the points are calculated for the multiple data points, and at the same time, based on the optimal clustering number k, the k-CAD is used to cluster the multiple data points to obtain the k-CAD value; Outlier abnormality degree assessment unit: assesses the abnormality degree of the outlier based on the peak point coordinates of the corresponding pulses on all frame segments on the xy coordinate axis; Classification and identification unit: classifies the detection signal according to the result of the clustering result calculation unit and the result of the outlier abnormality evaluation unit.
10. A method for clustering redundant object detection signals for aerospace sealed relay redundant object detection signals, characterized in that: The following steps are involved: Pulse extraction is performed on the PIND signal collected in the detection of redundant objects in aerospace relays, and the extracted pulse signal is framed; then each framed signal is used as a data point sample x i , convert each framed signal into an m-dimensional vector [x i1 ,x i2 ,…,x im ], use [x i1 ,x i2 ,…,x im ] as the data point x i The coordinates of ; all sets of data points are recorded as data set C: C={x i ∣i=1,2,…,n},x i =[x i1 ,x i2 ,…,x im ] Among them, x i represents a data point sample, [x i1 ,x i2 ,…,x im ] is x i The coordinates of , n is the number of data point samples; Based on the optimal number of clusters k, the k-means clustering algorithm is used to cluster the data set C to obtain the category label L = [l1, l2, ..., l n ], where l i Represents the signal point x i Category, l i ∈{1,2,…,k}; for each category j, j=1,2,…,k, extract all sample points belonging to this category, recorded as set C j ; The center μ of category j j The calculation formula is: Among them, |C j | is the number of sample points of category j; The k-CAD clustering result of category j is k-CAD(C j ) is used for the classification of redundant object detection signals of aerospace sealed relays.
Citation Information
Patent Citations
Based on improved fast density peak clustering and LOF outlier detection algorithm
CN109102028A
Power equipment discharge signal separation and classification method based on kernel principal component analysis
CN111444784A
Synchronous classification method for redundant object detection signals of spaceflight sealed electronic component
CN117009858A
Speed recognition method, system and device and storage medium
CN118226396A
PPG signal cluster center acquisition method and apparatus, and PPG signal processing method and apparatus
WO2023206888A1
