Method and device for predicting residual life of vacuum dry pump based on similarity matching

Abstract

The invention provides a vacuum dry pump residual life prediction method and device based on similarity matching, and relates to the technical field of equipment health monitoring. The method comprises the following steps: S1, acquiring vibration signal data of real-time operation monitoring of the vacuum dry pump; s2, performing feature preprocessing, data filtering and normalization on the vibration signal data, inputting a pre-trained vacuum dry pump residual life prediction model based on similarity matching, and outputting a prediction result; wherein the prediction result comprises the remaining service life of the vacuum dry pump at the fault occurrence moment; according to the vacuum dry pump residual life prediction model based on similarity matching, a local curve is intercepted through a distance measurement scale and window movement matching mode, similarity matching is carried out on the local curve and samples in a fault curve library, and therefore accurate prediction of the residual life of equipment is achieved. The method improves the quality of fault samples, reduces the number of samples, improves the prediction performance of the model, and reduces the calculation complexity and cost of the model.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment health monitoring, and in particular to a method and device for predicting the remaining life of a vacuum dry pump based on similarity matching. Background Art

[0002] The existing methods for predicting the remaining life of vacuum dry pumps are mainly divided into physical models, data-driven methods, and a combination of the two. The physical model method predicts the life of the equipment by establishing a mathematical model. Although it has high accuracy, it requires a lot of expert experience and prior knowledge and is difficult to adapt to complex systems. The data-driven method uses sensor data and machine learning technology for prediction. It has good versatility and adaptability, but may face challenges in data quality and labeling difficulties.

[0003] At present, the time series similarity method is one of the data-driven methods used by scholars in recent years to study the remaining life prediction. This method mainly uses a certain similarity evaluation scale to evaluate the similarity of the predicted time series and the historical time series through similarity matching, and then obtains the most similar historical time series for remaining life prediction. The similarity method is simple and intuitive, has high computational efficiency, and does not require complex model training, but it still depends on the number of fault samples. If the historical data is incomplete or biased, it may affect the accuracy of the prediction, especially in the selection of distance measurement scales. The measurement scales used by traditional methods are relatively common, and these methods have certain limitations in the accuracy of similarity matching.

[0004] Similarity matching methods and deep learning methods differ significantly in data processing and training prediction. Deep learning methods, such as the Transformer model, use complex neural network structures to capture complex patterns and long-term dependencies in data through multi-layer encoders and self-attention mechanisms. They are computationally expensive and require a large amount of training data. In contrast, similarity matching methods make predictions by marking fault features and comparing the similarity between current data and historical data. They do not require complex model training, are simple to calculate, and have higher accuracy.

[0005] In view of this, the applicant filed this application after studying the existing technology. Summary of the invention

[0006] The present invention aims to provide a method, device, equipment and medium for predicting the remaining life of a vacuum dry pump based on similarity matching, so as to solve the problems that the existing method is too dependent on fault samples and has low matching accuracy.

[0007] In order to solve the above technical problems, the present invention is implemented through the following technical solutions:

[0008] A method for predicting the remaining life of a vacuum dry pump based on similarity matching, comprising:

[0009] S1, obtaining vibration signal data of real-time operation monitoring of the vacuum dry pump;

[0010] S2, after feature preprocessing, data filtering and normalization of the vibration signal data, inputting a pre-trained vacuum dry pump remaining life prediction model based on similarity matching, and outputting a prediction result; wherein the prediction result includes the remaining service life of the vacuum dry pump at the time of the failure;

[0011] The vacuum dry pump remaining life prediction model based on similarity matching is trained by the following steps:

[0012] A1, obtain historical vibration signal data of downtime pumps and preventive downtime pumps;

[0013] A2, performing feature preprocessing on the historical vibration signal data to obtain time domain feature data;

[0014] A3, using the interquartile range method, selectively filtering the time domain characteristic data to obtain a time domain characteristic curve;

[0015] A4, dividing the time domain characteristic curve into a training curve library and a test curve library; sampling the time domain characteristic curve in the training curve library, intercepting the local fault curve that can characterize the occurrence of the fault, classifying the local fault curve according to the different fault curve change trends, and constructing a fault curve library; wherein each fault curve corresponds to the actual remaining service life of the vacuum dry pump;

[0016] A5, using the normalized training curve library and the fault curve library to train the remaining life prediction model, based on the distance measurement scale and the window moving matching method, a local curve is intercepted from each curve in the training curve library and normalized, and similarity matching is performed with the curves in the fault curve library; new fault samples are intercepted according to the similarity matching results, added to the fault curve library, and an updated fault curve library is obtained;

[0017] A6, input the normalized test curve library into the remaining life prediction model for testing until the model meets the evaluation criteria and obtains a trained vacuum dry pump remaining life prediction model based on the similarity method. Otherwise, retrain the model and construct the fault curve library.

[0018] Preferably, when the interquartile range method is used to perform selective filtering on the time domain feature data:

[0019] First, determine whether the data of any sampling point within the unit time length of each time domain characteristic data curve in the time domain characteristic data meets the filtering condition; if the filtering condition is met, the data of the sampling point is abnormal data, and the abnormal data is deleted;

[0020] Then, check whether there is a moment of missing data in the time domain feature data, and if so, use the value of the previous moment to fill it; if the moment of missing data appears at the beginning of the time domain feature data, use the median time period data of the time domain feature data to fill it;

[0021] The filtering condition is calculated based on the first quartile Q1 and the third quartile Q3 of the data in the unit time length, and the interquartile range IQR is calculated, IQR = Q3-Q1, so that the formula for the filtering condition is:

[0022] The time domain characteristic value is lower than Q1-K1*IQR, or higher than Q3+K2*IQR;

[0023] The first quartile Q1 is the value at the 25% position of the data within the unit time length after being arranged in ascending order; the third quartile Q3 is the value at the 75% position; K1 and K2 are weight coefficients respectively.

[0024] Preferably, the normalized formula is:

[0025]

[0026] Among them, x′ i is the normalized value of the ith data point of data X; X represents a continuous period of time domain feature data; x i is the value of the i-th data point of data X before normalization; max(X) and min(X) represent the maximum and minimum values ​​in data X, respectively.

[0027] Preferably, the distance metric is JS divergence, and the expression of JS divergence is:

[0028]

[0029] Among them, P and Q are two data sequences of the same length, and M is the average value of P and Q, that is,

[0030]

[0031] D KL (P||M) and D KL (Q||M) are the KL divergences of P and Q with respect to M respectively;

[0032]

[0033]

[0034] Wherein, i represents the i-th data of the data sequence, P(i) is the i-th data of the data sequence P, Q(i) is the i-th data of the data sequence Q, and M(i) is the i-th data of the data sequence M;

[0035] When performing similarity matching, if JS(P||Q) is less than the preset JS divergence threshold ε, the time domain feature data distribution similarity of P and Q is high; otherwise, the time domain feature data distribution similarity of P and Q is low.

[0036] Preferably, A5 is further:

[0037] Step 1: Set the size of the sliding window and the step length of the sliding window, use the window moving matching method to traverse each curve to be tested in the training curve library, move the window in the curve to be tested, and intercept the local curve; wherein the moving window of the curve to be tested is expressed as:

[0038] S t =Y[t,t+n-1], t=1,1+step,1+step*2,...,m-n+1;

[0039] Among them, S t represents the local curve at time t; Y is a continuous time domain characteristic data of the vacuum dry pump, i.e., the curve to be tested; t is the starting time of the sliding window; n is the size of the sliding window, which is equal to the length of the fault curve; step is the step length of the sliding window; m is the length of the curve data sequence to be tested;

[0040] Step 2: For each local curve S t Normalized and fault curve X in the fault curve library n Perform similarity matching to obtain the local curve S t The degree of similarity with each fault curve in the fault curve library;

[0041] Step 3: Select the first local curve S where the fault is detected in the curve Y to be tested. t =Y[t,t+n], and calculate the remaining service life of the curve to be tested at time t+n; the calculation formula for the remaining service life is:

[0042]

[0043] Where m represents the total number of sampling points of the curve Y to be measured, t ′ end Indicates the end time of the curve Y to be measured, that is, S t =Y[t ′ start ,t ′ end ], represents the local curve; is the curve Y to be measured at t′ end The actual remaining useful life at the moment;

[0044] The estimated remaining useful life is: RUL Y =RUL fault ;

[0045] Among them, RUL Y is the curve Y to be measured at t′ end The predicted remaining useful life at the time is equal to the local curve S t Remaining useful life RUL corresponding to similar failure curves fault .

[0046] Preferably, for each local curve S t Normalized and fault curve X in the fault curve library n When performing similarity matching, the degree of similarity is judged from both qualitative and quantitative aspects, specifically:

[0047] Qualitative judgment, based on the observation method, the extracted fault curve X n The local curve S intercepted by the curve to be measured t For comparison, if the local curve S t If the changing trend of the fault curve is similar to that in the existing fault curve library and conforms to the known fault category description, then it is qualitatively judged to be similar and belongs to the same category of fault;

[0048] Quantitative judgment, the fault curve X n With the local curve S t To calculate the distance metric, if S t If the distance metric calculated from X is less than the set threshold, it means that the curve to be tested is similar to the fault curve X and belongs to the corresponding fault category;

[0049] If the local curve S is qualitatively judged t With the fault curve X n The change trend of the local curve S is similar, and the distance measurement scale calculated during quantitative judgment is less than the set threshold, then the local curve S t With X n Belong to the same fault category;

[0050] If the local curve S is qualitatively judged t With the fault curve X n The change trend of is similar, but the distance measurement scale calculated in quantitative judgment is greater than the set threshold, then it is determined that S t With X n Belong to the same fault category, S tAdd it as a new fault curve to the fault samples of the corresponding category, and adjust the distance measurement scale threshold;

[0051] If the local curve S is qualitatively judged t With the fault curve X n If the change trends are not similar, but the distance measurement scale calculated in the quantitative judgment is less than the set threshold, the threshold of the distance measurement scale is adjusted;

[0052] If the local curve S is qualitatively judged t It has a clear trend of pump performance degradation, but it is different from the fault curve X n The change trends of are not similar, and the distance measurement scale calculated during quantitative judgment is greater than the set threshold, then it is determined that S t Does not belong to any of the existing fault categories, S t Add the fault curve library as a new fault type and adjust the distance metric threshold.

[0053] Preferably, when testing the model, each curve in the test curve library is traversed and matched with a similar local curve in the fault curve library to obtain the remaining service life of each curve in the test curve library at the time of the fault; the prediction error score is calculated according to the predicted remaining service life and the actual service life corresponding to each curve, and when the prediction error score is less than the set evaluation threshold, the model evaluation meets the standard; wherein the prediction error score is obtained by the accuracy rate, the false alarm rate and the missed alarm rate, specifically:

[0054] If the local curve S is found during qualitative judgment t Similar to the fault curve X, and quantitative judgment S t If the distance metric value calculated from X is less than the set threshold, the fault feature is correctly identified;

[0055] If the local curve S is found in qualitative judgment t Similar to the fault curve X, but when making quantitative judgments S t If the distance metric value calculated from X is greater than the set threshold, it is considered a false negative;

[0056] If the qualitative judgment is S t With fault sample S t Not similar, but when judging quantitatively the local curve S t If the distance measurement scale value calculated from the fault curve X is less than the set threshold, it is a false alarm;

[0057] If the local curve S is found during qualitative judgment t It is not similar to the fault curve X, and when quantitative judgment is made, S tIf the distance metric value calculated from X is greater than the set threshold, the fault feature is correctly identified;

[0058] The number of curves with correct similarity matching in the statistical test curve library is num T 、The number of all false positive curves is num FP 、The number of all missed curves is num FN , the total number of curves in the test curve library is num SUM ,but:

[0059] The accuracy formula is:

[0060] The formula for false alarm rate is:

[0061] The formula for the false negative rate is:

[0062] The prediction error score is used to evaluate the accuracy of the model prediction. The formula of the prediction error score S is:

[0063]

[0064] Where d is the prediction error of the curve to be tested, and RUL Y is the predicted remaining useful life of the curve to be tested, is the actual remaining service life of the curve to be tested;

[0065] If d>0, it means late prediction; if d<0, it means early prediction;

[0066] The smaller S is, the smaller the prediction error is and the more accurate the result is.

[0067] Preferably, the feature preprocessing includes performing time domain feature analysis on the vibration signal data, wherein the time domain feature analysis includes several of peak value, root mean square value, variance, skewness, kurtosis, energy, waveform factor and pulse factor.

[0068] Preferably, when performing time domain feature analysis, the abnormal performance of the fault curve includes a sharp rise in the curve, a sharp drop in the curve, a sudden change in the curve, and an oscillation in the curve.

[0069] The present invention also provides a vacuum dry pump remaining life prediction device based on similarity matching, comprising:

[0070] An acquisition unit, used for acquiring vibration signal data for real-time operation monitoring of a vacuum dry pump;

[0071] A prediction unit, used for performing feature preprocessing, data filtering and normalization on the vibration signal data, inputting a pre-trained vacuum dry pump remaining life prediction model based on similarity matching, and outputting a prediction result; wherein the prediction result includes the remaining service life of the vacuum dry pump at the time of the failure;

[0072] The remaining life prediction model of the vacuum dry pump based on similarity matching in the prediction unit is trained by the following steps:

[0073] A data sample acquisition module, used to obtain historical vibration signal data of downtime pumps and preventive downtime pumps;

[0074] A feature processing module, used for performing feature preprocessing on the historical vibration signal data to obtain time domain feature data;

[0075] A filtering module, used to selectively filter the time domain characteristic data using the interquartile range method to obtain a time domain characteristic curve;

[0076] A fault curve library construction module is used to divide the time domain characteristic curve into a training curve library and a test curve library; sample the time domain characteristic curve in the training curve library, intercept the local fault curve that can characterize the occurrence of the fault, classify the local fault curve according to the different fault curve change trends, and construct a fault curve library; wherein each fault curve corresponds to the actual remaining service life of the vacuum dry pump;

[0077] The matching training module is used to train the remaining life prediction model using the normalized training curve library and the fault curve library. Based on the distance measurement scale and the window moving matching method, a local curve is intercepted from each curve in the training curve library and normalized, and similarity matching is performed with the curves in the fault curve library. According to the similarity matching results, a new fault sample is intercepted and added to the fault curve library to obtain an updated fault curve library.

[0078] The test module is used to input the normalized test curve library into the remaining life prediction model for testing until the model meets the evaluation criteria and obtains a trained vacuum dry pump remaining life prediction model based on the similarity method. Otherwise, the model is retrained and the fault curve library is constructed.

[0079] The present invention also provides a vacuum dry pump remaining life prediction device based on similarity matching, comprising a processor and a memory, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement a vacuum dry pump remaining life prediction method based on similarity matching as described above.

[0080] The present invention also provides a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor of a device where the computer-readable storage medium is located, the method for predicting the remaining life of a vacuum dry pump based on similarity matching as described above is implemented.

[0081] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0082] The present invention pre-processes the vibration signal data of the vacuum dry pump, calculates the time domain features, and performs necessary filtering and normalization processing to improve the data quality. Then, the normalized time domain feature data is input into a pre-trained prediction model, which predicts the remaining service life of the pump based on historical data and similarity matching principles. In the model training stage, the time domain feature data is sampled, a fault curve library is constructed, and local curves are intercepted through normalization processing and moving matching methods, and similarity matching is performed with samples in the fault curve library, based on which the remaining life prediction value of the vacuum dry pump is calculated.

[0083] The present invention not only focuses on the operating status characteristics of the equipment, but also conducts in-depth analysis of the curve that characterizes the performance degradation trend of the vacuum dry pump, combines the distance measurement scale with the window moving matching method to perform similarity matching on the monitoring data curve, and effectively identifies whether the monitoring data has failed, thereby achieving accurate prediction of the remaining life of the equipment. The present invention uses the interquartile range method to filter the data and then normalizes it. The interquartile range method is not only used for outlier detection, but also for filling data gaps. It improves the quality of fault samples while reducing the number of fault samples, which not only improves the prediction performance of the model, but also reduces the computational complexity and cost of the model.

[0084] At the same time, the present invention uses JS divergence as an indicator for quantitative similarity judgment, processes one-dimensional dynamic time series data, and combines traditional qualitative judgment methods to make similarity judgments between the curve to be tested and the fault curve in the fault curve library, and applies it to the life prediction of vacuum dry pumps. By analyzing the time series data during the operation of the equipment, the changing trend of the equipment status is captured, thereby providing effective life prediction. This combined method has faster computing efficiency, does not require complex pre-training, reduces dependence on data volume, can quickly adapt to real-time changes in equipment status, effectively improves the accuracy of similarity matching, and provides a new and effective method for the maintenance and health management of mechanical equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0086] Figure 1 A schematic diagram of a method for predicting the remaining life of a vacuum dry pump based on similarity matching provided in Example 1 of the present invention.

[0087] Figure 2 A schematic diagram of training a vacuum dry pump remaining life prediction model based on similarity matching provided in Example 1 of the present invention.

[0088] Figure 3 This is a flow chart of the training of the remaining life prediction model for a vacuum dry pump provided in Example 1 of the present invention.

[0089] Figure 4 This is a flow chart of the test of the remaining life prediction model of a vacuum dry pump provided in Example 1 of the present invention.

[0090] Figure 5 A schematic diagram of a vacuum dry pump remaining life prediction device based on similarity matching provided in Embodiment 2 of the present invention.

[0091] Figure 6 This is a schematic diagram of the structure of a prediction unit provided in Embodiment 2 of the present invention.

[0092] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. DETAILED DESCRIPTION

[0093] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0094] Embodiment 1

[0095] Embodiment 1 of the present invention provides a method for predicting the remaining life of a vacuum dry pump based on similarity matching, which can be implemented by a vacuum dry pump remaining life prediction device based on similarity matching (hereinafter referred to as remaining life prediction), and in particular, is executed by one or more processors within the remaining life prediction.

[0096] In this embodiment, the remaining life prediction may be an electronic device equipped with a processor, the processor having a computer program of the vacuum dry pump remaining life prediction method based on similarity matching and the computer program can be executed, such as a computer, a smart phone, a smart tablet, a workstation, etc., which is not limited here.

[0097] In this embodiment, the similarity matching method and the deep learning method are significantly different in data processing and training prediction. Deep learning methods such as the Transformer model use a complex neural network structure to capture complex patterns and long-term dependencies in the data through multi-layer encoders and self-attention mechanisms. The computational cost is high and requires a large amount of training data. In contrast, the similarity matching method manually marks fault features and compares the similarity of current data with historical data to make predictions. It does not require complex model training, is simple to calculate, and has a higher accuracy rate.

[0098] Previous studies used the JS divergence formula for geometric distortion measurement, mainly processing photos and grayscale images taken by cameras, and evaluating the image adaptation quality by comparing pixels and their corresponding two-dimensional coordinates. This method is applicable to the field of image processing and focuses on the spatial analysis of static two-dimensional data. This patent processes one-dimensional dynamic time series data and is applied to the life prediction of vacuum dry pumps. By analyzing the time series data during the operation of the equipment, the changing trend of the equipment status is captured, thereby providing an effective life prediction.

[0099] In previous studies, the interquartile range (IQR) was mainly used for static data analysis, such as power metering error prediction. Its main function is to identify and process outliers in the data and improve the quality and accuracy of the data. This patent combines the interquartile range with the similarity matching model to optimize the data preprocessing process and applies it to the remaining life prediction of vacuum dry pumps in dynamic time series data. It is not only used for outlier detection, but also for filling data gaps, thereby improving the integrity and quality of the data.

[0100] like Figure 1 As shown, a method for predicting the remaining life of a vacuum dry pump based on similarity matching includes steps S1 to S2.

[0101] S1, obtaining vibration signal data of real-time operation monitoring of the vacuum dry pump;

[0102] S2, after feature preprocessing, data filtering and normalization of the vibration signal data, inputting a pre-trained vacuum dry pump remaining life prediction model based on similarity matching, and outputting a prediction result; wherein the prediction result includes the remaining service life of the vacuum dry pump at the time of the failure;

[0103] like Figure 2 and Figure 3 As shown, the vacuum dry pump remaining life prediction model based on similarity matching is obtained through training from steps A1 to A6:

[0104] A1, obtain the historical vibration signal data of the downtime pump and the preventive downtime pump.

[0105] In this embodiment, historical vibration signal data of the complete life cycle of the downtime vacuum dry pump from the start of operation to the shutdown is obtained, and the historical vibration signal data is time series data; historical vibration signal data of real-time monitoring of the vacuum dry pump that has been preventively shut down before shutdown is obtained, and the historical vibration signal data is time series data.

[0106] A2, performing feature preprocessing on the historical vibration signal data to obtain time domain feature data.

[0107] In this embodiment, based on the acquired historical vibration signal data, the time domain features that can reflect the basic statistical characteristics of the signal are calculated for the vibration data every second, and the time series data of the time domain features are obtained. Commonly used time domain features include but are not limited to peak value, root mean square value, variance, skewness, kurtosis, energy, waveform factor, pulse factor, etc.

[0108] Then, the calculated time domain features are analyzed to select those features that have obvious changing trends during the life cycle of the vacuum dry pump and are closely related to the health status and life of the pump. These key features should meet the following conditions:

[0109] The characteristic value has a significant change trend during the life cycle of the pump, such as gradual increase or decrease;

[0110] The change of characteristic value can reflect the performance attenuation and life status of the pump and has a high correlation.

[0111] A3, using the interquartile range method, selectively filtering the time domain feature data to obtain a time domain feature curve.

[0112] In an optional embodiment, step A3 specifically includes steps A31 to A34.

[0113] A31, respectively judge whether the data of any sampling point of each unit time length in a time domain characteristic data curve meets the filtering conditions; wherein the unit time length is M seconds; the filtering method adopted is the interquartile range method, and firstly, the first quartile (Q1) and the third quartile (Q3) of the data in the time period need to be calculated. Q1 is the value at the 25% position after the data in the unit time length are sorted from small to large, and Q3 is the value at the 75% position.

[0114] Next, the interquartile range IQR is calculated, and the formula is IQR = Q3-Q1. According to the interquartile range, the upper and lower thresholds of the filtering conditions are set, so that the formula for the filtering conditions is:

[0115] The time domain characteristic value is lower than Q1-K1*IQR, or higher than Q3+K2*IQR.

[0116] In this embodiment, assuming K1=K2=1.5, the filtering condition is: within each unit time length of the time domain characteristic data curve, for any sampling point, its time domain characteristic value is lower than Q1-1.5*IQR or higher than Q3+1.5*IQR.

[0117] A32, when the data of the sampling point meets the filtering condition, it is determined that the data contains abnormal data and the data segment is deleted.

[0118] A33, after deleting the abnormal data, check whether there are any moments with missing data in the time domain feature data. For the missing data points, use the value of the previous moment to fill them in to maintain the continuity of the data.

[0119] A34, if the data gap occurs at the beginning of the time domain characteristic data, the data in the time period is replaced by the median, thereby obtaining the filtered data of the downtime pump and the preventive downtime pump.

[0120] A4, dividing the time domain characteristic curve into a training curve library and a test curve library; sampling the time domain characteristic curve in the training curve library, intercepting the local fault curve that can characterize the occurrence of the fault, classifying the local fault curve according to the different fault curve change trends, and constructing a fault curve library; wherein each fault curve corresponds to the actual remaining service life of the vacuum dry pump.

[0121] In an optional embodiment, step A4 specifically includes steps A41 to A44.

[0122] A41, divides the time domain characteristic curve of the whole life cycle of the downtime pump into training curve and test curve; uses the real-time time domain characteristic curve of the preventive downtime pump as the test curve; combines the test curve of the downtime pump and the test curve of the preventive downtime pump into a test curve library.

[0123] A42, let the time domain characteristic curve after training curve library filtering be Y m ={y1,y2,...,y m}, where yi represents the time domain feature value of the i-th second in data Y, m represents the total number of seconds in Y, and y m is the time domain feature value of the mth second in the data Y.

[0124] Through analysis, local areas that can represent faults are identified in the time domain characteristic curve. Faults usually manifest as some abnormal patterns, such as sharp rise or fall, mutation, oscillation, etc. Specifically, the interval [t start ,t end ] to intercept the local data segment Y′={y tstart ,...,y tend}, and use it as the fault characteristic data segment X n ={x1,x2,...,x n}, that is, let X = Y′; where t start and t end Respectively represent the start and end time of the interval, y tstart is the time domain characteristic value at the beginning of the local data segment Y′, y tend is the time domain characteristic value at the end of the local data segment Y′, n represents the total number of seconds of the intercepted local data segment, and the calculation formula is n=t start -t end +1, and t start and t end Satisfy 1≤t start <t end ≤m, the fault data segment only accounts for a part of the total data, that is, n <m,x n It is the data value of the nth second of the intercepted fault feature data segment.

[0125] A43, for the intercepted fault characteristic data segment X n , calculate the corresponding remaining service life RUL fault , the calculation formula is: RUL fault =mt end , where RUL fault Indicates the remaining time from the end of the fault characteristic area to the pump shutdown or preventive shutdown.

[0126] A44, according to the different changing trends of the fault curves, the fault curves are divided into multiple categories, and all the intercepted and classified local fault curves are collected to form a fault curve library. There are different types of faults in the fault curve library, and each type of fault has multiple sample curves.

[0127] A5, the normalized training curve library and fault curve library are used to train the remaining life prediction model. Based on the distance measurement scale and window moving matching method, a local curve is intercepted from each curve in the training curve library and normalized, and similarity matching is performed with the curves in the fault curve library; new fault samples are intercepted according to the similarity matching results and added to the fault curve library to obtain an updated fault curve library.

[0128] In an optional embodiment, step A5 specifically includes steps A51 to A56.

[0129] A51, normalize the time domain characteristic data of each curve in the fault curve library, and use the normalized model to calculate each data point to obtain the normalized value.

[0130] A52, normalize the time domain feature data of each curve in the training curve library, and use the normalization model to calculate each data point to obtain a normalized value.

[0131] Wherein, the normalized model is:

[0132] Among them, x′ i is the normalized value of the ith data point of data X; X represents a continuous period of time domain feature data; x i is the value of the i-th data point of data X before normalization; max(X) and min(X) represent the maximum and minimum values ​​in data X, respectively.

[0133] A53, input the normalized training curve library and fault curve library into the remaining life prediction model for training, use the moving matching method to traverse each curve in the training curve library, set a sliding window, move the window on the time domain feature data of each curve, intercept the local curve, and normalize the local curve to determine the degree of similarity between the local curve and the fault curve in the fault curve library.

[0134] Among them, the sliding window in the mobile matching method is expressed as:

[0135] S t =Y[t,t+n-1],t=1,1+step,1+step*2,...,m-n+1;

[0136] Among them, S t represents the local curve at time t; Y is a continuous period of vacuum dry pump time domain characteristic data; t is the starting time of the sliding window; n is the window size, which is equal to the length of the fault curve; step is the step size of the sliding window; m represents the length of the data sequence.

[0137] A54, using the distance metric scale, performs similarity matching on each curve in the training curve library and the fault samples in the fault curve library in turn.

[0138] In this embodiment, the distance measurement scale adopts JS divergence, and the formula is:

[0139]

[0140] In the formula, P and Q are two data sequences of the same length, and M is the average value of P and Q, that is, D KL (P||M) and D KL (Q||M), are the KL divergences of P and Q with respect to M, Among them, i represents the i-th data of the data sequence, P(i) is the i-th data of the data sequence P, Q(i) is the i-th data of the data sequence Q, and M(i) is the i-th data of the data sequence M.

[0141] When performing similarity matching, a predefined JS divergence threshold ε is set to determine whether the JS divergence JS(P||Q) of curves P and Q is less than the threshold. If JS(P||Q)<ε, it is considered that the time domain feature data distribution of P and Q is highly similar; if JS(P||Q)≥ε, it is considered that the time domain feature data distribution of P and Q is insufficient to represent similarity.

[0142] It should be noted in this embodiment that the distance measurement scale may also be replaced by other distance measurement scale methods such as Euclidean distance and Manhattan distance, which are not limited here.

[0143] A55, compared with the local curve S t Compared with the fault curve X in the existing fault curve library n , calculate S t and X n JS divergence to determine whether their changing trends are similar.

[0144] If the local curve S is qualitatively judged t With the existing fault curve X n If the changing trends of the local curves are similar, but the calculated JS divergence is greater than the set threshold, it is determined that the local curves belong to the same fault category, and they are added as new fault curves to the samples of this type of fault, and the JS divergence threshold is adjusted.

[0145] If the local curve S is qualitatively judged t It has obvious changing trend that indicates pump performance degradation and is consistent with the existing fault curve X nIf the changing trends of are not similar and the calculated JS divergence is greater than the set threshold, it is determined that the local curve does not belong to any existing fault category, and it is added to the fault curve library as a new fault type, and the JS divergence threshold is adjusted.

[0146] If the local curve S is qualitatively judged t With the existing fault curve X n If the change trend of is similar to that of , and the calculated JS divergence is less than the set threshold, it is determined that the local curve belongs to this type of fault.

[0147] Among them, for each local curve S t and the fault curve X in the fault curve library n When performing similarity matching, the degree of similarity is judged from both qualitative and quantitative aspects, specifically:

[0148] Qualitative judgment, based on the observation method, the extracted fault curve X n The local curve S intercepted by the curve to be measured t For comparison, if the local curve S t If the changing trend of the fault curve is similar to that in the existing fault curve library and conforms to the known fault category description, then it is qualitatively judged to be similar and belongs to the same category of fault;

[0149] Quantitative judgment, the fault curve X n With the local curve S t Perform JS divergence calculation, if S t If the distance metric calculated from X is less than the set threshold, it means that the curve to be tested is similar to the fault curve X and belongs to the corresponding fault category;

[0150] If the local curve S is qualitatively judged t With the fault curve X n The change trend of the local curve S is similar, and the distance measurement scale calculated during quantitative judgment is less than the set threshold, then the local curve S t With X n They belong to the same fault category and the similarity matching is correct;

[0151] If the local curve S is qualitatively judged t With the fault curve X n The change trend of is similar, but the distance measurement scale calculated in quantitative judgment is greater than the set threshold, then it is determined that S t With X n Belong to the same fault category, S t Add it as a new fault curve to the fault samples of the corresponding category;

[0152] If the local curve S is qualitatively judged t With the fault curve X nIf the change trends of are not similar, but the distance measurement scale calculated in quantitative judgment is less than the set threshold, the distance measurement scale threshold is adjusted;

[0153] If the local curve S is qualitatively judged t It has a clear trend of pump performance degradation, but it is different from the fault curve X n The change trends of are not similar, and the distance measurement scale calculated during quantitative judgment is greater than the set threshold, then it is determined that S t Does not belong to any of the existing fault categories, S t Add the fault curve library as a new fault type and adjust the distance metric threshold.

[0154] A56, select a local curve S where the fault is first detected in the curve to be tested. t =Y[t,t+n], calculate the remaining life of the curve to be tested at time t+n.

[0155] The calculation formula for the actual remaining useful life is: Where m represents the total number of sampling points of Y, t′ end Represents the local curve S t =Y[t′ start ,t′ end ] At the end time of the curve Y to be measured, is the curve Y to be measured at t′ end The actual remaining useful life at the moment.

[0156] Predicted remaining useful life RUL Y =RUL fault , where RUL Y is the curve Y to be measured at t′ end The predicted remaining useful life at the time is equal to the local curve S t Remaining useful life RUL corresponding to similar failure curves fault .

[0157] A6, input the normalized test curve library into the remaining life prediction model for testing until the model meets the evaluation criteria and obtains a trained vacuum dry pump remaining life prediction model based on the similarity method. Otherwise, retrain the model and construct the fault curve library.

[0158] like Figure 4 As shown, in an optional embodiment, step A6 specifically includes steps A61 to A65.

[0159] A61, input the normalized test curve library into the trained remaining life prediction model, and use the moving matching and JS divergence methods to find the local curve similar to the fault curve on the time domain feature data of each curve in the test curve library, and then obtain the remaining service life of each curve to be tested in the test curve library at the time of the fault.

[0160] A62, based on the similarity matching results of each curve to be tested, calculate the accuracy, false alarm rate, and missed alarm rate, specifically:

[0161] If the local curve S is found during qualitative judgment t Similar to the fault curve X, and quantitative judgment S t If the JS divergence value calculated with X is less than the set threshold, the fault feature is correctly identified;

[0162] If the local curve S is found in qualitative judgment t Similar to the fault curve X, but when making quantitative judgments S t If the JS divergence value calculated with X is greater than the set threshold, it is a false negative;

[0163] If the qualitative judgment is S t With fault sample S t Not similar, but when judging quantitatively the local curve S t If the distance measurement scale value calculated from the fault curve X is less than the set threshold, it is a false alarm;

[0164] If the local curve S is found during qualitative judgment t It is not similar to the fault curve X, and when quantitative judgment is made, S t If the JS divergence value calculated with X is greater than the set threshold, the fault feature is correctly identified, that is, the curve to be tested is a healthy curve;

[0165] The number of curves with correct similarity matching in the statistical test curve library is num T 、The number of all false positive curves is num FP 、The number of all missed curves is num FN , the total number of curves in the test curve library is num SUM ,but:

[0166] The accuracy formula is:

[0167] The formula for false alarm rate is:

[0168] The formula for the false negative rate is:

[0169] A63, based on the predicted remaining useful life and the actual useful life corresponding to each curve to be tested, a prediction error score is calculated. When the prediction error score is less than a set evaluation threshold, the model evaluation meets the standard.

[0170] Furthermore, the formula for the prediction error score S is:

[0171]

[0172] Where d is the prediction error of the curve to be tested, and RUL Y is the predicted remaining useful life of the curve to be tested, is the actual remaining service life of the curve to be tested;

[0173] If d>0, it means late prediction; if d<0, it means early prediction;

[0174] The smaller S is, the smaller the prediction error is and the more accurate the result is.

[0175] A64, through the accuracy rate, false alarm rate and missed alarm rate of similarity matching, judge whether the fault curve library meets the standards. If the model is qualified, the trained vacuum dry pump remaining life prediction model based on the similarity method is obtained. Otherwise, the fault curve library is rebuilt until the model contains all the fault curves that can characterize the performance degradation of the pump.

[0176] Specifically, when the accuracy rate, false alarm rate, missed alarm rate of similarity matching and the accuracy rate of remaining life prediction (i.e., prediction error score S) are lower than the preset value, the model meets the requirements. The preset value is set by those skilled in the art according to actual needs, and the present invention does not specifically limit this.

[0177] In another preferred embodiment, in order to verify the effectiveness of the model and model solution proposed in the present invention, the historical vacuum dry pump operation data of a certain process condition of a certain company is selected as the research object. The data information includes 26 key time domain characteristic curves of 7 downtime pumps and 66 key time domain characteristic curves of 18 preventive downtime pumps, a total of 92 curves to be tested. The curves of 7 downtime pumps are selected as training curves, and the curves of 19 downtime pumps and 66 curves of preventive downtime pumps are selected as training curves, that is, a total of 7 curves are used as training curves for extracting fault curves, and a total of 85 curves are used as test curves.

[0178] For the training curve of the downtime pump, based on existing experience, local fault curves that can characterize the occurrence of faults are manually extracted. According to the different changing trends of the fault curves, they are divided into multiple categories to build a fault curve library, and each fault curve corresponds to its actual remaining service life.

[0179] The experimental parameters of the present invention are set as shown in Table 1 below:

[0180] Table 1. Experimental parameter settings

[0181] symbol meaning Example Value n The length of the intercepted fault curve 100000 seconds <![CDATA[RUL fault ]]> Remaining service life corresponding to the intercepted failure curve 1000000 seconds Size The size of the sliding window 100000 seconds Step The step size of the sliding window 1000 seconds

[0182] The similarity method of the present invention is used to train and test the model, and its performance effect is observed. The similarity matching results of the curves to be tested are shown in Table 2.

[0183] Table 2. Similarity matching results

[0184] The number of curves to be tested that are correctly matched by similarity 64 The number of curves to be tested with similarity matching false positives 18 The number of curves to be tested that are missed by similarity matching 3

[0185] According to the similarity matching results, we have:

[0186] num SUM =85;num T =64; num FP =18;num FN =3;

[0187] Therefore, the accuracy, false alarm rate and missed alarm rate of similarity matching are calculated and shown in Table 3.

[0188] Table 3. Similarity calculation results

[0189] Similarity matching accuracy TR 0.753 Similarity matching false alarm rate FPR 0.212 Similarity matching false negative rate FNR 0.035

[0190] Therefore, for the remaining life prediction results, among the 64 test curves with correct similarity matching, 63 test curves correctly predicted the remaining life.

[0191] Therefore, the accuracy of the remaining life prediction, that is, the prediction error score S, is 0.67.

[0192] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0193] The remaining life prediction model of the vacuum dry pump provided by the present invention not only focuses on the operating status characteristics of the equipment, but also can conduct in-depth analysis of the curve that characterizes the performance degradation trend of the vacuum dry pump. The model of the present invention can effectively identify whether the monitoring data has a fault by performing similarity matching on the monitoring data curve, thereby realizing accurate prediction of the remaining life of the equipment. The present invention improves the quality of the fault curve and reduces the number of fault curves by improving the data filtering preprocessing and normalization processing methods, which not only improves the prediction performance of the model, but also reduces the complexity and computational cost of the model.

[0194] The present invention uses JS divergence as a similarity measurement index. Compared with the traditional method, the similarity method has obvious advantages in the accuracy of similarity matching. The model not only reduces the dependence on the amount of data, but also can accurately identify the local curves in the curve to be tested that are highly similar to the fault curve, thus providing a new and effective method for the maintenance and health management of mechanical equipment.

[0195] Embodiment 2

[0196] like Figure 5 As shown, the second embodiment of the present invention further provides a vacuum dry pump remaining life prediction device based on similarity matching, comprising:

[0197] An acquisition unit, used for acquiring vibration signal data for real-time operation monitoring of a vacuum dry pump;

[0198] A prediction unit, used for performing feature preprocessing, data filtering and normalization on the vibration signal data, inputting a pre-trained vacuum dry pump remaining life prediction model based on similarity matching, and outputting a prediction result; wherein the prediction result includes the remaining service life of the vacuum dry pump at the time of the failure;

[0199] Specifically, Figure 6 As shown, the remaining life prediction model of the vacuum dry pump based on similarity matching in the prediction unit is trained by the following steps:

[0200] A data sample acquisition module, used to obtain historical vibration signal data of downtime pumps and preventive downtime pumps;

[0201] A feature processing module, used for performing feature preprocessing on the historical vibration signal data to obtain time domain feature data;

[0202] A filtering module, used to selectively filter the time domain characteristic data using the interquartile range method to obtain a time domain characteristic curve;

[0203] A fault curve library construction module is used to divide the time domain characteristic curve into a training curve library and a test curve library; sample the time domain characteristic curve in the training curve library, intercept the local fault curve that can characterize the occurrence of the fault, classify the local fault curve according to the different fault curve change trends, and construct a fault curve library; wherein each fault curve corresponds to the actual remaining service life of the vacuum dry pump;

[0204] The matching training module is used to train the remaining life prediction model using the normalized training curve library and the fault curve library. Based on the distance measurement scale and the window moving matching method, a local curve is intercepted from each curve in the training curve library and normalized, and similarity matching is performed with the curves in the fault curve library. According to the similarity matching results, a new fault sample is intercepted and added to the fault curve library to obtain an updated fault curve library.

[0205] The test module is used to input the normalized test curve library into the remaining life prediction model for testing until the model meets the evaluation criteria and obtains a trained vacuum dry pump remaining life prediction model based on the similarity method. Otherwise, the model is retrained and the fault curve library is constructed.

[0206] Embodiment 3

[0207] The third embodiment of the present invention also provides a vacuum dry pump remaining life prediction device based on similarity matching, which includes a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement the vacuum dry pump remaining life prediction method based on similarity matching as described above.

[0208] Embodiment 4

[0209] The fourth embodiment of the present invention further provides a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor of a device where the computer-readable storage medium is located, the method for predicting the remaining life of a vacuum dry pump based on similarity matching as described above is implemented.

[0210] In several embodiments provided in the embodiments of the present invention, it should be understood that the disclosed apparatus and method can also be implemented in other ways. The apparatus and method embodiments described above are merely schematic. For example, the flowcharts in the accompanying drawings show the possible architecture, functions and operations of the apparatus, method and computer program product according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0211] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0212] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention is essentially or partly contributed to the prior art or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, an electronic device, or a network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk. It should be noted that, in this article, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also includes elements inherent to such a process, method, article or device. Without more constraints, an element defined by the phrase "comprising a..." does not exclude the existence of other identical elements in the process, method, article or apparatus comprising the element.

[0213] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0214] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0215] The word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0216] The "first\second" mentioned in the embodiments is only to distinguish similar objects, and does not represent a specific order for the objects. It is understandable that the "first\second" can be interchanged with the specific order or sequence where permitted. It should be understood that the objects distinguished by "first\second" can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than those illustrated or described herein.

[0217] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting the remaining life of a vacuum dry pump based on similarity matching, characterized in that: include: S1, obtaining vibration signal data of real-time operation monitoring of the vacuum dry pump; S2, after feature preprocessing, data filtering and normalization of the vibration signal data, inputting a pre-trained vacuum dry pump remaining life prediction model based on similarity matching, and outputting a prediction result; wherein the prediction result includes the remaining service life of the vacuum dry pump at the time of the failure; The vacuum dry pump remaining life prediction model based on similarity matching is trained by the following steps: A1, obtain historical vibration signal data of downtime pumps and preventive downtime pumps; A2, performing feature preprocessing on the historical vibration signal data to obtain time domain feature data; A3, using the interquartile range method, selectively filtering the time domain characteristic data to obtain a time domain characteristic curve; A4, dividing the time domain characteristic curve into a training curve library and a test curve library; sampling the time domain characteristic curve in the training curve library, intercepting the local fault curve that can characterize the occurrence of the fault, classifying the local fault curve according to the different fault curve change trends, and constructing a fault curve library; wherein each fault curve corresponds to the actual remaining service life of the vacuum dry pump; A5, using the normalized training curve library and the fault curve library to train the remaining life prediction model, based on the distance measurement scale and the window moving matching method, a local curve is intercepted from each curve in the training curve library and normalized, and similarity matching is performed with the curves in the fault curve library; new fault samples are intercepted according to the similarity matching results, added to the fault curve library, and an updated fault curve library is obtained; A6, input the normalized test curve library into the remaining life prediction model for testing until the model meets the evaluation criteria and obtains a trained vacuum dry pump remaining life prediction model based on the similarity method. Otherwise, retrain the model and construct the fault curve library.

2. A method for predicting the remaining life of a vacuum dry pump based on similarity matching according to claim 1, characterized in that ,When the interquartile range method is used to selectively filter the time domain feature data: First, determine whether the data of any sampling point within the unit time length of each time domain characteristic data curve in the time domain characteristic data meets the filtering condition; if the filtering condition is met, the data of the sampling point is abnormal data, and the abnormal data is deleted; Then, check whether there is a moment of missing data in the time domain feature data, and if so, use the value of the previous moment to fill it; if the moment of missing data appears at the beginning of the time domain feature data, use the median time period data of the time domain feature data to fill it; The filtering condition is calculated based on the first quartile Q1 and the third quartile Q3 of the data in the unit time length, and the interquartile range IQR is calculated, IQR = Q3-Q1, so that the formula for the filtering condition is: The time domain characteristic value is lower than Q1-K1*IQR, or higher than Q3+K2*IQR; The first quartile Q1 is the value at the 25% position of the data within the unit time length after being arranged in ascending order; the third quartile Q3 is the value at the 75% position; K1 and K2 are weight coefficients respectively.

3. A method for predicting the remaining life of a vacuum dry pump based on similarity matching according to claim 2, characterized in that , the normalized formula is: Among them, x′ i is the normalized value of the ith data point of data X; X represents a continuous period of time domain feature data; x i is the value of the i-th data point of data X before normalization; max(X) and min(X) represent the maximum and minimum values ​​in data X, respectively.

4. The method for predicting the remaining life of a vacuum dry pump based on similarity matching according to claim 1 is characterized in that ,The distance measurement scale is JS divergence, and the expression of JS divergence is: Among them, P and Q are two data sequences of the same length, and M is the average value of P and Q, that is, D KL (P||M) and D KL (Q||M) are the KL divergences of P and Q with respect to M respectively; Wherein, i represents the i-th data of the data sequence, P(i) is the i-th data of the data sequence P, Q(i) is the i-th data of the data sequence Q, and M(i) is the i-th data of the data sequence M; When performing similarity matching, if JS(P||Q) is less than the preset JS divergence threshold ε, the time domain feature data distribution similarity of P and Q is high; otherwise, the time domain feature data distribution similarity of P and Q is low.

5. The method for predicting the remaining life of a vacuum dry pump based on similarity matching according to claim 3 is characterized in that , the A5 is further: Step 1: Set the size of the sliding window and the step length of the sliding window, use the window moving matching method to traverse each curve to be tested in the training curve library, move the window in the curve to be tested, and intercept the local curve; wherein the moving window of the curve to be tested is expressed as: S t =Y[t,t+n-1],t=1,1+step,1+step*2,...,m-n+1; Among them, S t represents the local curve at time t; Y is a continuous time domain characteristic data of the vacuum dry pump, i.e., the curve to be tested; t is the starting time of the sliding window; n is the size of the sliding window, which is equal to the length of the fault curve; step is the step length of the sliding window; m is the length of the curve data sequence to be tested; Step 2: For each local curve S t Normalized and fault curve X in the fault curve library n Perform similarity matching to obtain the local curve S t The degree of similarity with each fault curve in the fault curve library; Step 3: Select the first local curve S where the fault is detected in the curve Y to be tested. t =Y[t,t+n], and calculate the remaining service life of the curve to be tested at time t+n; the calculation formula for the remaining service life is: Where m represents the total number of sampling points of the curve Y to be measured, t′ end Indicates the end time of the curve Y to be measured, that is, S t =Y[t′ start ,t′ end ], represents the local curve; is the curve Y to be measured at t′ end The actual remaining useful life at the moment; The estimated remaining useful life is: RUL Y =RUL fault ; Among them, RUL Y is the curve Y to be measured at t′ end The predicted remaining useful life at the time is equal to the local curve S t Remaining useful life RUL corresponding to similar failure curves fault .

6. A method for predicting the remaining life of a vacuum dry pump based on similarity matching according to claim 5, characterized in that , for each local curve S t Normalized and fault curve X in the fault curve library n When performing similarity matching, the degree of similarity is judged from both qualitative and quantitative aspects, specifically: Qualitative judgment, based on the observation method, the extracted fault curve X n The local curve S intercepted by the curve to be measured t For comparison, if the local curve S t If the changing trend of the fault curve is similar to that in the existing fault curve library and conforms to the known fault category description, then it is qualitatively judged to be similar and belongs to the same category of fault; Quantitative judgment, the fault curve X n With the local curve S t To calculate the distance metric, if S t If the distance metric calculated from X is less than the set threshold, it means that the curve to be tested is similar to the fault curve X and belongs to the corresponding fault category; If the local curve S is qualitatively judged t With the fault curve X n The change trend of the local curve S is similar, and the distance measurement scale calculated during quantitative judgment is less than the set threshold, then the local curve S t With X n Belong to the same fault category; If the local curve S is qualitatively judged t With the fault curve X n The change trend of is similar, but the distance measurement scale calculated in quantitative judgment is greater than the set threshold, then it is determined that S t With X n Belong to the same fault category, S t Add it as a new fault curve to the fault samples of the corresponding category, and adjust the distance measurement scale threshold; If the local curve S is qualitatively judged t With the fault curve X n If the change trends are not similar, but the distance measurement scale calculated in the quantitative judgment is less than the set threshold, the threshold of the distance measurement scale is adjusted; If the local curve S is qualitatively judged t It has a clear trend of pump performance degradation, but it is different from the fault curve X n The change trends of are not similar, and the distance measurement scale calculated during quantitative judgment is greater than the set threshold, then it is determined that S t Does not belong to any of the existing fault categories, S t Add the fault curve library as a new fault type and adjust the distance metric threshold.

7. A method for predicting the remaining life of a vacuum dry pump based on similarity matching according to claim 6, characterized in that When testing the model, traverse and match each curve in the test curve library with the similar local curve in the fault curve library to obtain the remaining service life of each curve in the test curve library at the time of the fault; calculate the prediction error score according to the predicted remaining service life and the actual service life corresponding to each curve. When the prediction error score is less than the set evaluation threshold, the model evaluation meets the standard; wherein, the prediction error score is obtained by the accuracy rate, false alarm rate and missed alarm rate, specifically: If the local curve S is found during qualitative judgment t Similar to the fault curve X, and quantitative judgment S t If the distance metric value calculated from X is less than the set threshold, the fault feature is correctly identified; If the local curve S is found in qualitative judgment t Similar to the fault curve X, but when making quantitative judgments S t If the distance metric value calculated from X is greater than the set threshold, it is considered a false negative; If the qualitative judgment is S t With fault sample S t Not similar, but when judging quantitatively the local curve S t If the distance measurement scale value calculated from the fault curve X is less than the set threshold, it is a false alarm; If the local curve S is found during qualitative judgment t It is not similar to the fault curve X, and when quantitative judgment is made, S t If the distance metric value calculated from X is greater than the set threshold, the fault feature is correctly identified; The number of curves with correct similarity matching in the statistical test curve library is num T 、The number of all false positive curves is num FP 、The number of all missed curves is num FN , the total number of curves in the test curve library is num SUM ,but: The accuracy formula is: The formula for false alarm rate is: The formula for the false negative rate is: The prediction error score is used to evaluate the accuracy of the model prediction. The formula of the prediction error score S is: Where d is the prediction error of the curve to be tested, and RUL Y is the predicted remaining useful life of the curve to be tested, is the actual remaining service life of the curve to be tested; If d>0, it means late prediction; if d<0, it means early prediction; The smaller S is, the smaller the prediction error is and the more accurate the result is.

8. The method for predicting the remaining life of a vacuum dry pump based on similarity matching according to claim 1 is characterized in that ,The feature preprocessing includes time domain feature analysis of the vibration signal data, wherein the time domain feature analysis includes several of peak value, root mean square value, variance, skewness, kurtosis, energy, waveform factor and pulse factor.

9. The method for predicting the remaining life of a vacuum dry pump based on similarity matching according to any one of claim 1, characterized in that ,When conducting time domain feature analysis, the abnormal performance of the ,fault curve includes: sharp rise of the curve, sharp drop of the curve, sudden change of the curve and ,curve oscillation.

10. A vacuum dry pump remaining life prediction device based on similarity matching, characterized in that: include: An acquisition unit, used for acquiring vibration signal data for real-time operation monitoring of a vacuum dry pump; A prediction unit, used for performing feature preprocessing, data filtering and normalization on the vibration signal data, inputting a pre-trained vacuum dry pump remaining life prediction model based on similarity matching, and outputting a prediction result; wherein the prediction result includes the remaining service life of the vacuum dry pump at the time of the failure; The remaining life prediction model of the vacuum dry pump based on similarity matching in the prediction unit is trained by the following steps: A data sample acquisition module, used to obtain historical vibration signal data of downtime pumps and preventive downtime pumps; A feature processing module, used for performing feature preprocessing on the historical vibration signal data to obtain time domain feature data; A filtering module, used to selectively filter the time domain characteristic data using the interquartile range method to obtain a time domain characteristic curve; A fault curve library construction module is used to divide the time domain characteristic curve into a training curve library and a test curve library; sample the time domain characteristic curve in the training curve library, intercept the local fault curve that can characterize the occurrence of the fault, classify the local fault curve according to the different fault curve change trends, and construct a fault curve library; wherein each fault curve corresponds to the actual remaining service life of the vacuum dry pump; The matching training module is used to train the remaining life prediction model using the normalized training curve library and the fault curve library. Based on the distance measurement scale and the window moving matching method, a local curve is intercepted from each curve in the training curve library and normalized, and similarity matching is performed with the curves in the fault curve library. According to the similarity matching results, a new fault sample is intercepted and added to the fault curve library to obtain an updated fault curve library. The test module is used to input the normalized test curve library into the remaining life prediction model for testing until the model meets the evaluation criteria and obtains a trained vacuum dry pump remaining life prediction model based on the similarity method. Otherwise, the model is retrained and the fault curve library is constructed.

Images