A method and device for screening similar samples of a device degradation process

Abstract

The application discloses a kind of screening method and device of equipment degradation process similar sample, it is related to aerospace technical field, its method includes: first, the monitoring parameter data collected by target equipment and sensor of same type equipment are reduced parameter dimension by PCA, realize the effect of extracting main information of degradation;Then, according to the length of target equipment degradation period, determine the weight of bulldozer distance and cosine distance in integrated distance measurement process;Next, based on bulldozer distance and cosine distance measurement method and the weight determined by itself, the integrated distance between target equipment and each same type equipment is calculated, and the similarity between same type equipment and target equipment is represented by this. Finally, the similar same type equipment of target equipment is selected by two methods of sorting optimization and statistical screening in turn.

Description

Technical Field

[0001] This invention relates to the field of aerospace technology, and in particular to a method and apparatus for screening similar samples of equipment degradation processes. Background Technology

[0002] In aerospace, advanced equipment, and other fields, condition assessment and remaining capacity prediction models are indispensable parts of equipment use and maintenance. However, when using data-driven assessment and prediction, the target equipment (the equipment being assessed and predicted) has relatively short historical operating data, and its performance degradation patterns are not yet clearly apparent, making it difficult to effectively build a data-driven model. This poses a challenge to accurate equipment condition assessment and prediction. Utilizing a large amount of end-of-life operating data collected from other equipment of the same model is a possible solution. However, in reality, different equipment of the same model generally have differences in manufacturing deviations, initial failure levels, and degradation rates, resulting in significant differences in sensor monitoring data and performance degradation trends. Sensor data collected from other equipment cannot be directly used to effectively establish an assessment and prediction model for the target equipment, thus affecting the accuracy of the results.

[0003] The amount of shared information between equipment of the same model and the target equipment constitutes sample similarity. Similarity measurement allows for the selection of highly similar equipment and data samples from a database of equipment of the same model. Existing similarity measurement methods include Euclidean distance, Mahalanobis distance, Chebyshev distance, and cosine distance. However, these methods primarily focus on the numerical distribution of the data and do not address the characteristics of equipment degradation process data. The main drawbacks of existing technologies are:

[0004] 1) Sample similarity was not calculated by taking into account the degradation rate during equipment operation;

[0005] 2) The issue of different initial degradation levels and different degradation periods among different devices was not taken into account. Summary of the Invention

[0006] To address the aforementioned problems, embodiments of the present invention provide a method and apparatus for screening similar samples in the equipment degradation process.

[0007] A method for screening similar samples of equipment degradation process according to an embodiment of the present invention includes:

[0008] Obtain the degradation period length of the target device and the multidimensional monitoring parameters related to degradation. At the same time, obtain M devices of the same model as the target device and obtain the multidimensional monitoring parameters related to degradation for each device of the same model.

[0009] The first main degradation information was extracted from the multidimensional monitoring parameters of the target device using principal component analysis, and the second main degradation information was extracted from the multidimensional monitoring parameters of each device of the same model.

[0010] Based on the degradation period length of the target equipment, the cosine distance weight and the bulldozer distance weight are calculated respectively.

[0011] Using the first degradation main information, the second degradation main information, the cosine distance weight, and the bulldozer distance weight, the similarity value between the target equipment and each of the same type of equipment is calculated;

[0012] Using the similarity values ​​between the target device and each of the same model devices, N similar devices of the same model are selected from the M devices of the same model;

[0013] Wherein, M and N are both positive integers, and M is greater than N.

[0014] A device for screening similar samples of equipment degradation process according to an embodiment of the present invention includes:

[0015] The acquisition module is used to acquire the degradation period length of the target device and the multidimensional monitoring parameters related to degradation, and at the same time acquire M devices of the same model as the target device, and acquire the multidimensional monitoring parameters related to degradation for each device of the same model.

[0016] The extraction module is used to extract the first main degradation information from the multidimensional monitoring parameters of the target device using principal component analysis, and to extract the second main degradation information from the multidimensional monitoring parameters of each device of the same model.

[0017] The calculation module is used to calculate the cosine distance weight and the bulldozer distance weight according to the degradation period length of the target equipment; and to calculate the similarity value between the target equipment and each of the same model equipment using the first degradation main information, the second degradation main information, the cosine distance weight and the bulldozer distance weight.

[0018] The filtering module is used to filter out N similar devices from the M similar devices using the similarity value between the target device and each of the same model devices;

[0019] Wherein, M and N are both positive integers, and M is greater than N.

[0020] According to the solution provided in this embodiment of the invention, it is possible to screen for similar equipment with a similar degradation trajectory to the target equipment and their corresponding sample data locations from existing full-lifecycle sample data sets of similar equipment by combining the initial degradation degree and degradation rate. This provides high-quality sample data for performance status assessment and remaining capacity prediction of the target equipment, thereby improving the effectiveness of assessment and prediction. Attached Figure Description

[0021] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to understand the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0022] Figure 1 This is a flowchart of a method for screening similar samples in a device degradation process provided by an embodiment of the present invention;

[0023] Figure 2 This is a schematic diagram of a screening device for similar samples in the equipment degradation process provided in an embodiment of the present invention;

[0024] Figure 3 This is a flowchart of the time-delay dynamic integrated distance metric sample similarity metric method provided in the embodiments of the present invention;

[0025] Figure 4 This is a schematic diagram of the integrated distance dynamic weight provided in an embodiment of the present invention;

[0026] Figure 5 This is a schematic diagram of TLDW-EDM similarity measurement provided in an embodiment of the present invention;

[0027] Figure 6 This is a PCA result diagram of the same model engine provided in an embodiment of the present invention;

[0028] Figure 7 This is a PCA result diagram of the target engine provided in an embodiment of the present invention;

[0029] Figure 8 This is a schematic diagram of the target engine similarity sample screening results provided in an embodiment of the present invention. Detailed Implementation

[0030] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the preferred embodiments described below are only for illustration and explanation of the present invention and are not intended to limit the present invention.

[0031] This invention addresses the characteristics of equipment degradation by integrating two indicators: numerical distribution and degradation rate. It proposes a weighted ensemble measurement method combining bulldozer distance and cosine distance. Bulldozer distance identifies samples with a health status numerical distribution consistent with the target equipment's degradation trajectory, while cosine distance identifies samples with a rate (angle) consistent with the target equipment's degradation trajectory. Furthermore, this invention employs a time-delay distance measurement method to address the issue of varying initial fault levels among samples. A time-delay-integrated similarity measurement method is proposed to calculate sample similarity and filter out similar equipment and data samples of the same model with similar degradation processes to the target equipment, providing stronger data support for equipment condition assessment and prediction.

[0032] To support condition-based maintenance, users often need sufficient data for data-driven condition monitoring, performance evaluation, and fault diagnosis. In the data selection process, the similarity to the target equipment's operating state significantly impacts data usability. Therefore, this invention proposes a sample similarity measurement and screening method based on Time Lagand Dynamic Weight Ensemble Distance Measure (TLDW-EDM) to screen sample data with high similarity to the target equipment. Sample data can include various operational monitoring parameters or environmental monitoring parameters collected during equipment operation. Samples with high similarity have two main characteristics: similar parameter values ​​and similar parameter degradation rates. Represented in the data, this translates to differences in absolute values ​​and rates of change. To integrate these two evaluation indicators, this invention uses a weighted integration method based on bulldozer distance and cosine distance to measure the similarity between data samples from the same model of equipment and data samples from the target equipment. Simultaneously, this invention employs a time-delay distance measurement method to address the issue of different initial fault levels among samples. For the target equipment, this method measures the sample similarity of the same model of equipment, and similar samples are determined through a two-stage screening process. The process is as follows: Figure 3 As shown, the input data for the method of this invention originates from multi-dimensional monitoring parameter data collected by sensors installed on the equipment or in the environment during equipment operation. First, the monitoring parameter data collected by sensors on the target equipment and similar equipment are processed using PCA (Principal Component Analysis) to reduce the dimensionality of the parameters, thereby extracting key degradation information. Then, the weights of bulldozer distance and cosine distance in the integration distance measurement process are determined based on the length of the target equipment's degradation period. Next, the integration distance between the target equipment and each similar equipment is calculated based on the bulldozer distance and cosine distance measurement methods and their respective determined weights, and this distance represents the similarity between the similar equipment and the target equipment. Finally, similar equipment of the same model is selected sequentially through both ranking and statistical screening methods.

[0033] Figure 1 This is a flowchart of a method for screening similar samples in a device degradation process provided by an embodiment of the present invention, as shown below. Figure 1 As shown, it includes:

[0034] Step S101: Obtain the degradation period length of the target device and the multidimensional monitoring parameters related to degradation. At the same time, obtain M devices of the same model as the target device and obtain the multidimensional monitoring parameters related to degradation for each device of the same model.

[0035] Step S102: Extract the first main degradation information from the multidimensional monitoring parameters of the target device using principal component analysis, and extract the second main degradation information from the multidimensional monitoring parameters of each device of the same model;

[0036] Step S103: Calculate the cosine distance weight and bulldozer distance weight according to the degradation period length of the target equipment;

[0037] Step S104: Using the first degradation main information, the second degradation main information, the cosine distance weight, and the bulldozer distance weight, calculate the similarity value between the target device and each device of the same model;

[0038] Step S105: Using the similarity value between the target device and each of the same model devices, select N similar devices from the M devices of the same model that are similar to the target device;

[0039] Wherein, M and N are both positive integers, and M is greater than N.

[0040] Furthermore, the step of calculating the cosine distance weight and the bulldozer distance weight based on the degradation period length of the target equipment includes: the degradation period length of the target equipment is directly proportional to the cosine distance weight and inversely proportional to the bulldozer distance weight.

[0041] Specifically, the calculation of the cosine distance weight and the bulldozer distance weight based on the degradation period length of the target equipment includes:

[0042]

[0043]

[0044] Among them, W cos The weights for the cosine distance; t l T represents the length of the degradation period of the target device. up ,T lower Let T be the upper and lower bounds of the distance weights, respectively. up +T lower =1; Lmax ,L min These represent the maximum and minimum values ​​of the weight change length range, respectively; W EMD The weight of the bulldozer distance; and W cos +W EMD =1.

[0045] Further, the step of calculating the similarity value between the target device and each device of the same model using the first degradation main information, the second degradation main information, the cosine distance weight, and the bulldozer distance weight includes: obtaining a target degradation feature sequence from the first degradation main information, and obtaining a same-model degradation feature sequence from the second degradation main information; using the target degradation feature sequence and the same-model degradation feature sequence, calculating the cosine distance and bulldozer distance between the target degradation feature sequence and the same-model degradation feature sequence, respectively; and calculating the similarity value between the target device and each device of the same model based on the cosine distance, the bulldozer distance, the cosine distance weight, and the bulldozer distance weight.

[0046] Specifically, calculating the similarity value between the target device and each device of the same model based on the cosine distance, the bulldozer distance, the cosine distance weight, and the bulldozer distance weight includes: performing time-delay segmentation on the full-lifetime degradation trajectory of each device of the same model based on the degradation period length of the target device, obtaining multiple degradation feature sub-sequences of the same model for each device of the same model; using a time-delay distance metric method, calculating the integrated distance between each degradation feature sub-sequence of the same model for each device of the same model and the target degradation feature sequence, obtaining an integrated distance set for each device of the same model, and using the minimum integrated distance value in the integrated distance set as the similarity value; wherein, calculating the integrated distance between each degradation feature sub-sequence of the same model for each device of the same model and the target degradation feature sequence includes:

[0047] d t =W cos *d cos (Z s Z t ,t)+W EMD *d EMD (Z s Z t ,t)

[0048] Where, d t W represents the integration distance at time delay t. cos , where W is the cosine distance weight; EMD For the bulldozer distance weight, d cos d is the cosine distance; EMDZ represents the distance to the bulldozer. s These are characteristic sequences of the same type of degradation; Z t t represents the target degradation feature sequence; t is the starting point of the time delay sliding window.

[0049] Further, selecting N similar devices from the M similar devices using the similarity values ​​between the target device and each of the same model includes: statistically analyzing the similarity values ​​between the target device and each of the same model to obtain M similarity values; sorting the M similarity values ​​in ascending order and selecting the top Y similar devices as pre-screened similar devices; and using a similar sample secondary screening method to perform secondary screening on the pre-screened similar devices to select N similar devices from the Y pre-screened similar devices; wherein Y is a positive integer, and M is greater than Y and N.

[0050] The process of using a similar sample secondary screening method to perform secondary screening on the pre-screened similar devices to select N similar devices to the target device includes: obtaining the start time corresponding to the highest similarity sequence of each pre-screened similar device and the full life cycle of the pre-screened similar device; calculating the remaining usage time of each pre-screened similar device based on the start time and the full life cycle; calculating the mean and variance of the remaining usage time of each pre-screened similar device based on the remaining usage time, and constructing a remaining usage time interval for secondary screening of similar devices using the mean and variance of the remaining usage time; comparing the remaining usage time of each pre-screened similar device with the remaining usage time interval, and identifying pre-screened similar devices whose remaining usage time falls within the remaining usage time interval as similar devices to the target device.

[0051] Specifically, calculating the remaining usage time of each pre-screened device of the same model, based on the start time and the full life cycle, includes:

[0052] t j * =arg min t (D j )

[0053]

[0054] In the formula, The starting time corresponding to the sequence with the highest similarity to device j of the same model is used for pre-screening; D j RUL is a set of similar distances for pre-screening devices j of the same model; j *To pre-screen the remaining usage time of equipment j of the same model; t Ej This is to pre-screen the entire life cycle of equipment j of the same model.

[0055] Figure 2 This is a schematic diagram of a screening device for similar samples in the equipment degradation process provided in an embodiment of the present invention, as shown below. Figure 2 As shown, the system includes: an acquisition module 201, used to acquire the degradation period length of the target device and the degradation-related multidimensional monitoring parameters, and simultaneously acquire M devices of the same model as the target device, and acquire the degradation-related multidimensional monitoring parameters of each device of the same model; an extraction module 202, used to extract the first main degradation information from the multidimensional monitoring parameters of the target device using principal component analysis, and extract the second main degradation information from the multidimensional monitoring parameters of each device of the same model; a calculation module 203, used to calculate the cosine distance weight and the bulldozer distance weight according to the degradation period length of the target device; and to calculate the similarity value between the target device and each device of the same model using the first main degradation information, the second main degradation information, the cosine distance weight, and the bulldozer distance weight; and a filtering module 204, used to filter out N devices of the same model similar to the target device from the M devices of the same model using the similarity value between the target device and each device of the same model; wherein M and N are both positive integers, and M is greater than N.

[0056] The degradation period of the target equipment is directly proportional to the cosine distance weight and inversely proportional to the bulldozer distance weight.

[0057] Figure 3 This is a flowchart of the time-delay dynamic integrated distance metric sample similarity metric method provided in the embodiments of the present invention, as follows: Figure 3 As shown, it includes:

[0058] 1) Extraction of master information on equipment degradation based on principal component analysis

[0059] Principal component analysis is a statistical method for dimensionality reduction. It uses orthogonal transformation to convert variables that may be correlated into mutually independent variables, thus constructing new characteristic variables. This set of transformed variables is called principal components.

[0060] There is some information redundancy among the equipment monitoring parameters. To reduce computational load and error accumulation, principal component analysis (PCA) is used to extract the main information from the monitoring parameters, reducing data dimensionality and achieving information reduction and parameter dimensionality reduction. This method takes multi-parameter monitoring data of the target equipment and similar equipment as input. Through orthogonal transformation, correlated parameters are orthogonally transformed into independent variables. The variable with the largest contribution is output as the degradation principal component, which is the equipment degradation curve—a one-dimensional curve. This reduces data dimensionality and volume, avoids the impact of information redundancy on prediction accuracy, and also reduces computational costs.

[0061] 2) Dynamic weighted integrated distance metric method based on bulldozer distance and cosine distance

[0062] Once equipment reaches a specific performance standard during operation, it is considered to have transitioned from a healthy operating state to a degraded operating state. The duration of this degraded operating state is defined as the length of the degradation period. The dynamic weighted integrated distance metric method uses the length of the target equipment's degradation period as input. Different target equipment have different degradation period lengths, and the distribution range is quite large. For shorter degradation features, they are easily affected by acquisition errors and operating environments during the degradation process, resulting in limited representation of the degradation rate. Therefore, greater attention should be paid to the similarity of the original data values ​​of the monitoring parameters. For longer degradation features, the degradation trajectory can better represent the corresponding degradation rate. In this case, compared to the original values, the degradation rate feature is more representative of the equipment degradation process, and greater attention should be paid to the similarity of the degradation rates. This invention selects two distance metric methods—bulldozner distance and cosine distance—to represent the original parameter values ​​and degradation rate, respectively, as similarity evaluation indicators. Figure 4 As shown, for target equipment with a shorter degradation period, the cosine distance weight is set to be smaller, while the bulldozer distance weight is set to be larger. For individuals with a longer degradation period, the weight settings are reversed.

[0063] To prevent excessive weighting of any distance from causing an imbalance in the similarity index, this invention sets upper and lower thresholds for the weights of each distance. When the degradation period of the target device is t... l When the weights of the two distances are calculated, the formulas are as follows:

[0064]

[0065]

[0066] In the formula T up ,T lower These are the upper and lower bounds of the two distance weights, respectively, and T up +T lower =1. L max ,L min These represent the maximum and minimum values ​​of the weight change range, t.l Let W be the degradation period length of the target equipment. The above formula allows for the adaptive determination of the weights W between the bulldozer distance and the cosine distance for different target equipment. cos and W EMD W cos +W EMD =1. By using two distance weight values, two distance metrics can be combined to improve the accuracy of similarity measurement.

[0067] 3) Similarity measurement method based on TLDW-EDM

[0068] Due to individual manufacturing errors and other factors, the initial fault levels of different devices vary. To better measure the similarity between samples of the same model and the target device, this invention designs the TLDW-EDM method, which adds a sliding window delay process to the dynamic integrated distance metric. The degradation trajectory length t of the target device is used as the reference. l Given the window length, the degradation trajectory of the same model equipment throughout its entire lifespan is time-delayed and segmented to obtain the segmented data. The segmented data is the data segment from time t to time t+tl in the degradation feature sequence Zs of the same model. This segment of data has the same length as the target degradation feature sequence Zt, which is convenient for distance calculation.

[0069] like Figure 5 As shown in the figure, the curves represent the degradation characteristics of samples of the same model, the hollow circular curves represent the degradation characteristics of the target domain, and the solid circular curves represent the degradation characteristics of the target domain after time delay. When the starting point of the time delay sliding window is t, the ending point is t+t. l . t E For the entire lifecycle of the same model sample, Z s Zs is the extracted master information, which is the degradation feature sequence of the same model. Since the data for the same model of equipment is relatively long, the extracted master information is denoted as the degradation feature sequence Zs. It should be noted that Zs represents the entire degradation trajectory of the same model of equipment throughout its lifespan, and different values ​​of t correspond to a subset of Zs that is cut out and is of the same length as Zt. t Let Zt be the target degradation feature sequence (Zt is the extracted master information; the target device data is relatively short, so the extracted master information is denoted as the target degradation feature sequence Zt), and d be the distance between the two sequences. The formula for calculating the time delay distance metric is as follows:

[0070] d t =W cos *d cos (Z s Z t ,t)+W EMD *d EMD (Z s Z t ,t)

[0071] In the formula W cos W EMD d represents the weights of the cosine distance and the bulldozer distance, respectively. cos d EMD The cosine distance and bulldozer distance are respectively represented by d. t It is the variable-weighted integration distance between the target sample and the sample of the same model at time t, that is, the distance between the device of the same model and the target device to be predicted, at the time t delay position corresponding to the data segment.

[0072] For vector Z s =(x1,x2,x3........x n ) and Z t =(y1,y2,y3.......y n In an n-dimensional vector space, their cosine similarity is:

[0073]

[0074] The bulldozer distance is defined as follows:

[0075]

[0076] In the formula, Π(Z) s Z t ) is a distribution Z s Z t The set of all possible joint distributions. Samples x and y are collected from each possible joint distribution, and the distance between these pairs is calculated; this gives the expected distance of the sample pairs under that possible joint distribution. The lower bound of the expected distance obtained across all possible joint distributions is the bulldozer distance.

[0077] As the delay start point t changes, the remaining usage time value at the end of the window also changes. The formula for calculating the remaining usage time corresponding to each delay start point t is as follows:

[0078] RUL t =t E -(t+t l )

[0079] In the formula, t E For the entire lifecycle of the same model sample, t l The length of the degradation trajectory of the target device. RUL t The remaining lifetime corresponds to time t, the starting point of the target degradation characteristic delay.

[0080] Through the time delay process, the similarity distance set formed by the target device and the device of the same model corresponding to different time delay cutting times t is obtained. Similarity is represented by the minimum value in the set of similar distances. That is:

[0081]

[0082] Where, γ j D represents the similarity of the same model of equipment with the number j. j This is the set of similar distances for devices of the same model.

[0083] 4) Similar Sample Screening Methods

[0084] Based on the results of similarity measurement, a similar sample screening method was designed to make full use of the information of samples of the same model. For the target device, the similar sample screening process is divided into sample pre-screening based on minimum similarity distance and sample secondary screening based on statistical methods.

[0085] 1) Sample pre-screening based on minimum similarity distance

[0086] If a segment of a degradation feature sequence from the same type of equipment shows a small dynamic weighted integration distance with the target degradation feature, the sample from that same type is considered to have high similarity. Therefore, during the pre-screening process, the similarity of all equipment of the same type is calculated based on the time-delay dynamic integration distance metric, and the equipment of the same type is sorted in ascending order of similarity. The top-ranked S-values ​​are then selected. tr The equipment was used as a sample for pre-screening.

[0087] 2) Secondary screening of similar samples

[0088] The pre-screened equipment samples have sample sequences that are quite similar to the target equipment. However, the target equipment has a relatively short operating time, resulting in occasional similarity among the samples. Therefore, to exclude accidentally similar equipment of the same model, this invention designs a secondary screening process for similar samples. Based on the pre-screened samples, the start time of each pre-screened sample is calculated. Corresponding Remaining Lifetime (RUL) j *

[0089] t j * =arg min t (D j )

[0090]

[0091] In the formula, D represents the start time corresponding to the sequence with the highest similarity to the same type of equipment. j This is a set of similar distances for devices of the same model. Remaining Usage Time (RUL) j *T represents the lifetime corresponding to the sequence position with the highest similarity among devices j of the same model. Ej This refers to the total operating life of equipment j of the same model.

[0092] For a specific target device, the operating sequences of devices with similar degradation processes exhibit a relatively uniform remaining lifetime distribution. Therefore, a secondary screening is performed on the initially selected samples of the same model based on the remaining lifetime (RUL*). The mean μ and variance σ of the remaining lifetime (RUL*) label values ​​for the samples of the same model are calculated. The RUL* interval for the secondary screening is [μ-r·σ, μ+r·σ], where r is the secondary screening factor. Samples whose remaining lifetime (RUL*) labels are not within the screening range are eliminated in the secondary screening, and the remaining samples are the final similar samples.

[0093] Example

[0094] This invention utilizes the Data Challenge dataset from the 2008 PHM International Conference for methodological research and validation. This data was obtained through C-MAPSS (Commercial Modular Aero-Propulsion System Simulation). The dataset contains simulation data from 100 aero-engines. Through parameter selection, eight gas path parameters related to degradation were chosen for case studies.

[0095] This invention uses PCA to reduce the dimensionality of multi-dimensional engine airflow parameters. Eight engine parameters are input into a PCA model, and three principal components within these eight dimensions are analyzed. The results show that after principal component extraction, the first principal component (PCA) contributes 85% of the total, containing most of the degradation information from the original parameter values. Therefore, the first PCA is selected as the primary source of engine degradation information to remove redundant information from the monitoring parameters and reduce computational costs. The dataset includes PCA analysis results for 100 engines of the same model and the target engine, as shown below. Figure 6-7 As shown.

[0096] Using the time-delay dynamic weighted integrated distance metric method proposed in this invention, for a target engine, the similarity of 100 samples of the same engine model is measured, and training samples with high similarity are selected. The parameter settings are shown in Table 1. The variable weight range of the integrated metric based on bulldozer distance and cosine distance is set to [1 / 3, 2 / 3], that is, 1 / 3 of the weight is bulldozer distance, 1 / 3 of the weight is cosine distance, and the remaining 1 / 3 of the distance weight is determined by the degradation period length of the target engine. According to the statistics of the degradation period length of the same engine model, the maximum and minimum degradation period lengths are 31 and 362, respectively. Therefore, L max =362, L min=31, the target engine degradation period length is 73. The weights of the cosine distance and bulldozer distance are determined to be 37.56% and 62.44% respectively through dynamic weight calculation.

[0097] The TLDW-EDM method proposed in this invention is used to screen similarity samples from 100 engines of the same model. This invention sets the initial screening sample size S for the target engines. tr The value is set to 10, meaning that the top 10% of engines with the highest similarity among the source domain samples are selected as the initial screening samples. The secondary screening factor r is set to 1.5, retaining samples with remaining usage time in the interval [μ-1.5σ, μ+1.5σ] as secondary screening samples.

[0098] Table 1 Sample Similarity Measurement and Screening Parameters

[0099] Parameter name <![CDATA[T up ]]> <![CDATA[T lower ]]> <![CDATA[S tr ]]> r Parameter value 1 / 3 2 / 3 10 1.5

[0100] The proposed method initially screened out engines of the same model that showed high similarity to the target engine and matched the sample sequences with the highest similarity. The preliminary screening results of the target engine are as follows: Figure 8 As shown in -a. The results show that the similarity measurement method proposed in this invention can screen out engines of the same model and sample locations that have similar degradation characteristic trajectories to the target test engine. The results of a secondary screening of the selected engines based on the remaining usage time of the initially screened samples are shown in... Figure 8 As shown in -b. The results show that as similarity decreases, the distance between the engine samples of the same model and the target engine samples increases, and the distribution of remaining usage time corresponding to the samples gradually becomes unstable. Through secondary screening, the results are as follows... Figure 8 As shown in -c, engine samples with deviations in remaining usage time distribution are removed, and the similar samples corresponding to the target engine are finally determined. The screening results are shown in Table 2 below.

[0101] Table 2: Screening Results Table

[0102] Same model engine number Pre-screening results 11、37、13、22、63、38、51、85、70、68 Secondary screening results 11、37、13、22、63、38、51、85、70

[0103] The solution provided by the embodiments of the present invention includes the following: 1) measuring sample distance from multiple aspects using two distance measurement methods to ensure the comprehensiveness of sequence decay feature screening; 2) adjusting the distance measurement weight according to the data length to reduce the measurement error caused by data fluctuation.

[0104] Although the present invention has been described in detail above, it is not limited thereto, and those skilled in the art can make various modifications based on the principles of the present invention. Therefore, all modifications made in accordance with the principles of the present invention should be understood to fall within the protection scope of the present invention.

Claims

1. A method for screening similar samples in a device degradation process, characterized in that, The method comprises the following steps: acquiring the degradation period length of a target device and multi-dimensional monitoring parameters related to degradation, and acquiring M devices of the same type as the target device and multi-dimensional monitoring parameters related to degradation of each device of the same type; extracting first degradation main information from the multi-dimensional monitoring parameters of the target device and second degradation main information from the multi-dimensional monitoring parameters of each device of the same type by a principal component analysis method; calculating a cosine distance weight and a bulldozer distance weight according to the degradation period length of the target device, which comprises: wherein, W cos is the weight of the cosine distance; t l is the length of the degradation period of the target device; T up , T lower are the upper and lower limits of the distance weight, respectively, and T up + T lower = 1; L max , L min are the maximum and minimum values of the weight change length range, respectively; W EMD is the weight of the bulldozer distance; and W cos + W EMD = 1; wherein the degradation period length of the target device is proportional to the cosine distance weight and inversely proportional to the bulldozer distance weight; calculating similarity values between the target device and each device of the same type by using the first degradation main information, the second degradation main information, the cosine distance weight and the bulldozer distance weight; screening N devices of the same type similar to the target device from the M devices of the same type by using the similarity values between the target device and each device of the same type; wherein M and N are positive integers, and M is greater than N.

2. The method of claim 1, wherein, The method of calculating the similarity values between the target device and each device of the same type by using the first degradation main information, the second degradation main information, the cosine distance weight and the bulldozer distance weight comprises: acquiring a target degradation feature sequence from the first degradation main information and a same-type degradation feature sequence from the second degradation main information; calculating a cosine distance and a bulldozer distance between the target degradation feature sequence and the same-type degradation feature sequence by using the target degradation feature sequence and the same-type degradation feature sequence; calculating the similarity values between the target device and each device of the same type according to the cosine distance, the bulldozer distance, the cosine distance weight and the bulldozer distance weight.

3. The method of claim 2, wherein, The method of calculating the similarity values between the target device and each device of the same type according to the cosine distance, the bulldozer distance, the cosine distance weight and the bulldozer distance weight comprises: delay-cutting the full-life degradation trajectory of each device of the same type according to the degradation period length of the target device to obtain a plurality of same-type degradation feature subsequences of each device of the same type; obtaining an integrated distance set of each device of the same type by calculating an integrated distance between each same-type degradation feature subsequence of each device of the same type and the target degradation feature sequence by using a time-delay distance measurement method, and taking the minimum integrated distance value in the integrated distance set as a similarity value; wherein the method of calculating the integrated distance between each same-type degradation feature subsequence of each device of the same type and the target degradation feature sequence comprises: wherein, d t is the integrated distance corresponding to the time delay t; W cos , is the cosine distance weight; W EMD is the bulldozer distance weight, d cos is the cosine distance; d EMD is the bulldozer distance; Z s is the same type of degradation feature sequence; Z t is the target degradation feature sequence; t is the starting point of the time delay sliding window.

4. The method of claim 3, wherein, screening N devices of the same type similar to the target device from the M devices of the same type by using the similarity values between the target device and each device of the same type. statistically obtaining M similarity values between the target device and each of the M same-model devices; arranging the M similarity values in ascending order, and selecting Y same-model devices with high ranks as pre-screened same-model devices; performing secondary screening on the pre-screened same-model devices by using a similar sample secondary screening method, and selecting N same-model devices similar to the target device from the Y pre-screened same-model devices; wherein Y is a positive integer, M is greater than Y, and Y is greater than N.

5. The method of claim 4, wherein, performing secondary screening on the pre-screened same-model devices by using a similar sample secondary screening method, and selecting N same-model devices similar to the target device include: obtaining a starting time corresponding to a highest sequence of similarity of each pre-screened same-model device and a full life cycle of the pre-screened same-model device, and calculating a remaining use time of the pre-screened same-model device according to the starting time and the full life cycle; calculating a mean value and a variance of the remaining use time according to the remaining use time of each pre-screened same-model device, and constructing a remaining use time interval for performing secondary screening on the same-model devices by using the mean value and the variance of the remaining use time; comparing the remaining use time of each pre-screened same-model device with the remaining use time interval, and regarding the pre-screened same-model device with the remaining use time located in the remaining use time interval as a same-model device similar to the target device.

6. The method of claim 5, wherein, calculating the remaining use time of each pre-screened same-model device according to the starting time and the full life cycle include: In the formula, is the start time corresponding to the highest similarity sequence of the same type of equipment j pre-screened; D j is the similarity distance set of the same type of equipment j pre-screened; RUL j * is the remaining useful life of the same type of equipment j pre-screened; t Ej is the life cycle of the same type of equipment j pre-screened.

7. A screening device for similar samples of a device degradation process, said device being arranged to implement a method according to any one of claims 1-6, characterized in that, include: an obtaining module, configured to obtain a degradation period length of a target device and a plurality of degradation-related monitoring parameters of the target device, and obtain M same-model devices of the same model as the target device and a plurality of degradation-related monitoring parameters of each same-model device; an extracting module, configured to extract first degradation main information from the plurality of monitoring parameters of the target device by using a principal component analysis method, and extract second degradation main information from the plurality of monitoring parameters of each same-model device; a calculating module, configured to calculate a cosine distance weight and a bulldozer distance weight respectively according to the degradation period length of the target device, and calculate a similarity value between the target device and each same-model device by using the first degradation main information, the second degradation main information, the cosine distance weight, and the bulldozer distance weight; a screening module, configured to select N same-model devices similar to the target device from the M same-model devices by using the similarity value between the target device and each same-model device; wherein M and N are positive integers, and M is greater than N.

8. The apparatus of claim 7, wherein, The degradation period length of the target device is proportional to the cosine distance weight and inversely proportional to the bulldozer distance weight.

Images