Future blind missing data completion method for real-time equipment health monitoring

Abstract

The present invention relates to the technical field of intelligent operation and maintenance of mechanical equipment, and specifically to a future blind missing data completion method for real-time health monitoring of equipment, comprising: reading real-time operating data of mechanical equipment; using an indicator to classify whether the real-time operating data is lost; constructing a distance measurement matrix to quantify the degree of correlation between the missing data and forward known data; constructing a hysteresis degradation function based on the distance measurement matrix; reading the autocorrelations of the forward known data, fitting the weight matrix and bias vector parameter values in the hysteresis degradation function; measuring the influence weight of the forward known data on the missing data through the hysteresis degradation function; constructing a monotonic matrix to determine the monotonicity of the forward known data; and completing the missing data based on the monotonic matrix of the forward known data and the influence weight of the forward known data on the missing data. The present invention can reduce the loss of degradation information and maintain the original degradation trajectory.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance of mechanical equipment, and more particularly to a future blind lost data completion method for real-time health monitoring of equipment. Background Art

[0002] Modern manufacturing is inseparable from mechanical manufacturing. As an important component of the manufacturing industry, mechanical equipment plays a vital role in improving production efficiency, reducing costs, and improving product quality. Predictive maintenance of mechanical equipment can avoid catastrophic failures, reduce maintenance costs, and effectively improve the safety, reliability, and economic efficiency of machines. Therefore, predictive maintenance of mechanical equipment is of great significance in the mechanical manufacturing industry. Predictive maintenance of mechanical equipment relies on real-time monitoring data for judgment. In actual industrial scenarios, observation noise, sensor failure, and other factors can lead to data loss, making it impossible to obtain complete monitoring data. The high data loss rate poses a significant obstacle to predictive maintenance of mechanical equipment. How to complete the missing data during real-time monitoring is a key research issue.

[0003] In the existing health monitoring process, no effective missing data completion method has been proposed. When missing data is encountered, the main option is direct elimination. This method will cause information loss and lead to deviations in the data distribution. It is only applicable to cases with a small missing ratio. When the missing ratio is high, direct elimination will lead to a large amount of degradation information loss, which will directly affect the prediction effect and reduce the prediction accuracy. In the field of data processing, a common method is the filling method. This method uses the statistical characteristics of the non-missing data before and after the missing data to fill the missing data. The filling results obtained by this method are stable and suitable for cases where the distribution characteristics of the missing sequence are simple and clear and the variables are highly correlated. The disadvantage of the filling method is that it requires the full life cycle operation data set for completion. That is, it needs to combine the forward and backward data relationships of the missing data for completion. However, real-time health monitoring can only obtain forward known data and cannot obtain data after the missing data, that is, future blind situations. Therefore, the missing data completion methods in the field of data processing cannot be directly applied to the predictive maintenance process of mechanical equipment.

[0004] Therefore, how to find a data completion method that can minimize the loss of mechanical equipment degradation information and have good completion performance when only forward data is known and a high data missing rate is an urgent problem that technicians in this field need to solve. Summary of the Invention

[0005] In view of this, the present invention provides a future blind lost data completion method for real-time health monitoring of equipment, which can reduce the loss of degradation information and maintain the original degradation trajectory.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a method for future blind missing data completion for real-time health monitoring of equipment, comprising the following steps:

[0008] Read real-time operation data of mechanical equipment;

[0009] Use indicators to classify whether real-time operation data is lost;

[0010] When missing data occurs, a distance metric matrix is constructed to quantify the degree of association between the missing data and the forward known data;

[0011] Construct a hysteresis degradation function based on the distance metric matrix to describe the degradation relationship between data;

[0012] Read the autocorrelations of the forward known data and fit the weight matrix and bias vector parameter values in the hysteresis degradation function;

[0013] The influence weight of forward known data on missing data is measured by the hysteresis degradation function;

[0014] Construct a monotonic matrix to determine the monotonicity of the forward known data;

[0015] The missing data is completed according to the monotone matrix of the forward known data and the influence weight of the forward known data on the missing data.

[0016] Furthermore, the indicator is a binary vector used to indicate whether data is lost. Indicates the Point-in-time data Lost, Indicates the Point-in-time data Not lost.

[0017] Furthermore, the distance metric matrix construction process includes:

[0018] when When the distance metric matrix The expression is: ;

[0019] when When the distance metric matrix The expression is: ;

[0020] in, is the last absolute time interval, The time position of lost data, is the position of the last data point that was not lost; is the missing pattern of data features at the previous moment; Recursion to , stop calculating the distance between data, The smaller it is, the closer the distance between the data in the time dimension is, and the greater the degree of correlation influence.

[0021] Furthermore, the autocorrelation of the forward preset number of unmissed running data is read through the autocovariance function. , hysteresis The autocorrelation function of order is expressed as:

[0022] ;

[0023] in, Lag The autocovariance function of order, is the zero-lag autocovariance function, is the number of running data, represents the selected number of time lags, for Running data value at all times, for Running data value at all times, is the average of the running data; combined with the missing data indicator , if and only if and At the same time, Only then will there be calculated value.

[0024] Furthermore, the expression of the hysteresis degradation function is:

[0025] ;

[0026] in, ; is the weight matrix, Bias vector is the parameter that needs to be fitted; represents the distance metric matrix.

[0027] Furthermore, the process of fitting the weight matrix and bias vector parameter values in the hysteresis degradation function includes:

[0028] In order to combine the data with the relevant information, , read the data degradation information, that is, use the following model to fit the unknown parameters:

[0029] ;

[0030] in, is the distance metric vector corresponding to the running data under the lag step;

[0031] Use square error to construct the loss function and calculate the weight matrix by minimizing the loss and the bias vector , the specific calculation formula is:

[0032] ;

[0033] The calculated weight matrix and the bias vector Substitute into the hysteresis degradation function.

[0034] Furthermore, the expression of the monotone matrix is:

[0035] ;

[0036] in, is the time point where the data is lost. If the overall running data is monotonically increasing, the monotonic matrix ; If the overall running data is monotonically decreasing, the monotonic matrix .

[0037] Furthermore, the calculation formula for completing the missing data is:

[0038] ;

[0039] in, For the The data after the missing data at the time point is supplemented, and Forward known running data and For lost data Impact weight; is the global average, that is, the average of all forward known data; , indicating the Data is lost at a certain point in time.

[0040] In a second aspect, the present invention provides a future blind lost data completion device for real-time health monitoring of equipment, comprising:

[0041] Data acquisition module, used to read real-time operating data of mechanical equipment;

[0042] A data loss classification module, for classifying whether the real-time running data is lost using an indicator;

[0043] The distance measurement module is used to construct a distance measurement matrix when missing data occurs to quantify the degree of correlation between the missing data and the forward known data;

[0044] The data correlation description module is used to construct a hysteresis degradation function based on the distance measurement matrix to describe the degradation relationship between data; read the autocorrelations of the forward known data, fit the weight matrix and bias vector parameter values in the hysteresis degradation function; and measure the influence weight of the forward known data on the missing data through the hysteresis degradation function;

[0045] Data monotonicity determination module, used to construct a monotonic matrix and determine the monotonicity of forward known data;

[0046] The data completion module is used to complete the missing data based on the monotone matrix of the forward known data and the influence weight of the forward known data on the missing data.

[0047] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the future blind lost data completion method for real-time health monitoring of equipment as described above are implemented.

[0048] It can be seen from the above technical solutions that compared with the prior art, the present invention has the following beneficial effects:

[0049] 1) The present invention quantifies the relationship between missing data and existing data by constructing a distance metric matrix, ensuring that the supplementary value does not affect the overall monotonicity of the mechanical equipment degradation data.

[0050] 2) The present invention reads the autocorrelation information of the forward known data by constructing an autocovariance function, and fits the position parameters of the lag degradation function in combination with the autocorrelation information, thereby calculating the influence weights of different forward known data on the missing data, and then completing the data in combination with the monotonicity of the forward known data. This process can capture the irregular lag relationship between the missing data and the forward observations to avoid losing valuable information. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0052] Figure 1 A flow chart of the future blind lost data completion method for real-time equipment health monitoring provided by the present invention;

[0053] Figure 2 This is the overall architecture diagram of the future blind lost data completion method for real-time equipment health monitoring provided by the present invention;

[0054] Figure 3 Schematic diagram of performance comparison between the present invention and the existing method. DETAILED DESCRIPTION

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0056] like Figure 1-Figure 2 As shown, the embodiment of the present invention discloses a future blind missing data completion method for real-time health monitoring of equipment, including the following steps:

[0057] S1. Read the real-time operation data of mechanical equipment;

[0058] S2. using an indicator to classify whether the real-time operation data is lost;

[0059] S3. When missing data occurs, a distance metric matrix is constructed to quantify the degree of correlation between the missing data and the forward known data;

[0060] S4. Constructing a hysteresis degradation function based on the distance metric matrix to describe the degradation relationship between data;

[0061] Read the autocorrelations of the forward known data and fit the weight matrix and bias vector parameter values in the hysteresis degradation function;

[0062] S5. Measure the influence weight of forward known data on missing data through hysteresis degradation function;

[0063] Construct a monotonic matrix to determine the monotonicity of the forward known data;

[0064] The missing data is completed according to the monotone matrix of the forward known data and the influence weight of the forward known data on the missing data.

[0065] The above steps are further explained below.

[0066] S1. Read the real-time operation data of mechanical equipment. Use sensors to obtain the real-time operation data of mechanical equipment. Any data can be expressed as ,in, Indicates the operation data of mechanical equipment. , is the dimension of the running data, For a certain point in time, Indicates the operating time of mechanical equipment.

[0067] S2. Classify the real-time operation data.

[0068] Setting the Lost Data Indicator , indicator is a binary vector used to indicate whether data is lost. Indicates the Point-in-time data Lost, Indicates the Point-in-time data If no data is lost, the data will be retained in the real-time running data.

[0069] S3. Construct a distance metric matrix. The specific construction process includes:

[0070] The operation data of mechanical equipment is time-varying. As the distance between data increases, the correlation between data will gradually decrease. In addition, in real situations, the missing position is random and unevenly distributed. Therefore, the distance metric matrix is introduced. The distance between the missing data and the non-missing data is used to reflect the degree of correlation between data points.

[0071] when When the distance metric matrix The expression is: ;

[0072] when When the distance metric matrix The expression is: ;

[0073] in, is the last absolute time interval, The time position of lost data, is the position of the last data point that was not lost; is the missing pattern of data features at the previous moment; Recursion to , stop calculating the distance between data, The smaller it is, the closer the distance between the data in the time dimension is, and the greater the degree of correlation influence.

[0074] S4. Obtain the temporal correlation between data, including:

[0075] S41. Read the autocorrelated information of the data that is not lost.

[0076] The operation data of mechanical equipment has autocorrelation. The autocovariance function is constructed to read the autocorrelation of the operation data. The operation data is grouped according to the location of the missing data. Each group consists of the missing data and the data of the previous ten time points. , hysteresis The autocorrelation function of order is expressed as:

[0077] ;

[0078] in, Lag The autocovariance function of order, is the zero-lag autocovariance function, is the number of running data, represents the selected number of time lags, for Running data value at all times, for Running data value at all times, is the average of the running data; combined with the missing data indicator , if and only if and At the same time, Only then will there be calculated value.

[0079] S42. Read the timing correlation between the lost data and the forward known data.

[0080] In order to read the temporal correlation between the non-lost data and the known data, maintain the monotonicity of the original data, and reduce the loss of degradation information, the hysteresis degradation function is introduced. Describe the degradation relationship between data, the expression is:

[0081] ;

[0082] in, ; is the weight matrix, The bias vector is the parameter that needs to be fitted. and Use the autocorrelation function results for fitting; represents the distance metric matrix.

[0083] S43. Parameter fitting and weight calculation.

[0084] In order to combine the data autocorrelation information, is the influence weight of the forward known data on the actual data, and , read the data degradation information, that is, use the following model to fit the unknown parameters:

[0085] ;

[0086] in, is the distance metric vector corresponding to the running data under the lag step;

[0087] Using squared error Construct a loss function, minimize the loss, use the least squares method to fit, and calculate the weight matrix and the bias vector , the specific calculation formula is:

[0088] ;

[0089] The calculated weight matrix and the bias vector Substitute into the hysteresis degradation function.

[0090] S5. Complete the missing data based on the forward known data.

[0091] S51. Read the monotonicity of the mechanical equipment operation data.

[0092] Constructing a monotone matrix , read the monotonicity of the forward known data to avoid the trend of the original running data being changed by the completed data. The specific expression is:

[0093] ;

[0094] in, is the time point where the data is lost. If the overall running data is monotonically increasing, the monotonic matrix ; If the overall running data is monotonically decreasing, the monotonic matrix .

[0095] Missing data may also occur in the early stages of operation. When the second data point is missing, the monotonic matrix value cannot be determined using only the first data point. The range of 3 to 5 is set because when the missing data point is between 3 and 5, there are at least 2 forward known data points, and only 2 data points can be used for determination. The range of 5 to 7 is set because there are only 4 forward known data points. The more known data points, the easier it is to determine the monotonicity of the running data. However, too many data points are meaningless for determining monotonicity. The main focus is on the data near the missing data. Therefore, the maximum number of data points used is set to 6.

[0096] S52, combining the hysteresis degradation function value obtained above to characterize the time series correlation and missing data indicator , fill in the missing data, by lagging the degradation function Measuring missing data With forward known data and The degree of influence between and , the missing data completion model is:

[0097] ;

[0098] in, For the The data after the missing data at the time point is supplemented, and Forward known running data and For lost data Impact weight; is the global average, that is, the average of all forward known data; , indicating the Data is lost at a certain point in time.

[0099] The method of the present invention is a real-time missing data completion method. During the real-time monitoring process, when the first missing data point is encountered, the missing data completion mechanism is triggered. After the completion is complete, the completed data is returned to the known data set, and real-time monitoring continues until the next missing data is encountered.

[0100] When data is missing continuously, the last completed data is classified as known data. That is, after completing the first data, the first completed data is identified as known data, and then the next missing data is completed. As a known data, combined with Perform data completion.

[0101] To avoid continuous loss The calculation result is too small to affect the correct summation calculation, increase the global Additional supplementation is performed and finally weighted summation is used to complete the missing data.

[0102] S6. Real-time monitoring of running data. When missing data is encountered again, the next round of missing data completion is performed, and the above steps are repeated until all real-time monitoring data completion is completed.

[0103] In other embodiments, the present invention further provides a future blind lost data completion device for real-time health monitoring of equipment, comprising:

[0104] Data acquisition module, used to read real-time operating data of mechanical equipment;

[0105] A data loss classification module, for classifying whether the real-time running data is lost using an indicator;

[0106] The distance measurement module is used to construct a distance measurement matrix when missing data occurs to quantify the degree of correlation between the missing data and the forward known data;

[0107] The data correlation description module is used to construct a hysteresis degradation function based on the distance measurement matrix to describe the degradation relationship between data; read the autocorrelations of the forward known data, fit the weight matrix and bias vector parameter values in the hysteresis degradation function; and measure the influence weight of the forward known data on the missing data through the hysteresis degradation function;

[0108] Data monotonicity determination module, used to construct a monotonic matrix and determine the monotonicity of forward known data;

[0109] The data completion module is used to complete the missing data based on the monotone matrix of the forward known data and the influence weight of the forward known data on the missing data.

[0110] In another embodiment, the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the future blind missing data completion method for real-time health monitoring of equipment are implemented as described above.

[0111] Next, taking the trend prediction of rail transit bearings as an example, the effectiveness of the method of the present invention was verified through experimental bench prediction simulation data.

[0112] The test bearings used in this experiment were HRB 352213 double-row tapered roller bearings. The bearing parameters and fault characteristic frequencies are shown in Table 1. The test bearings in the experimental rig were belt-driven by a drive motor (a three-phase AC asynchronous motor). Axial and radial loads on the test bearings were applied to the bearing supports via a hydraulic loading device. This example was a constant-operation experiment, with a motor speed of 2500 rpm, an axial load of -20 kN, and a radial load of 60 kN. A triaxial accelerometer was used to monitor the test and support bearings. A current clamp was used to monitor the three-phase current of the motor, and a speed sensor was used to measure the speed (rotational frequency) in real time. The sampling strategy for the bearing vibration operating signal was set as follows: a sampling frequency of 25.6 kHz, a sampling interval of 30 seconds, and a sampling duration of 3.84 seconds, meaning each sample contained 98,304 data points.

[0113] Table 1. HRB 352213 geometric parameters and their fault characteristic frequencies

[0114]

[0115] Table 2 shows the actual operating life and failure mode of each test bearing.

[0116] Table 2. Test bench bearing accelerated degradation test information

[0117]

[0118] The experiment uses the method proposed in the present invention to complete the bearing operation data. The experiment compares the method proposed in the present invention with three common methods: KNN, forward filling, and Random Forest. Before completing the missing data, the bearing operation data are grouped to simulate the data loss in real scenarios. Six missing patterns are randomly set for each group of data. The specific missing scenario settings are shown in Table 3.

[0119] Table 3. Missing scene settings

[0120]

[0121] The different missing rates of the running data are compared, Table 4 and Figure 3 The comparative effects of prognostic difference and mean relative error of three bearing operation data with five missing rates are listed.

[0122] Table 4 Comparison of average relative errors of four methods

[0123]

[0124] The experimental results show that as the missing rate increases, the completion effect of the method of the present invention is significantly better than the other three comparison methods, while better retaining the degradation information of the bearing.

[0125] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0126] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A future blind missing data completion method for real-time equipment health monitoring, characterized by: The following steps are involved: Read real-time operation data of mechanical equipment; Use indicators to classify whether real-time operation data is lost; When missing data occurs, a distance metric matrix is constructed to quantify the degree of association between the missing data and the forward known data; The construction process of the distance metric matrix includes: When l t-1 =0,l t =1, the expression of the distance metric matrix ψ is: t =|M t -M t-1 |; When l t-1 =1,l t =1, the expression of the distance metric matrix ψ is: t =|M t -M t-1 |+ψ t-1 ; Among them, ψ t-1 is the last absolute time interval, M t is the time position of lost data, M t-1 is the position of the last data point that was not lost; l t-1 is the data feature missing pattern of the previous moment; t =|M t -M t-1 |+ψ t-1 Recursion to l t-k =0, stop calculating the distance between data, ψ t The smaller the value, the closer the distance between data in the time dimension, and the greater the degree of correlation influence; A hysteresis degradation function is constructed based on the distance metric matrix to describe the degradation relationship between data. The expression of the hysteresis degradation function is: Among them, θ∈[0,1]; a is the weight matrix, b is the bias vector, which is the parameter to be fitted; ψ represents the distance metric matrix; Read the autocorrelations of the forward known data and fit the weight matrix and bias vector parameter values in the hysteresis degradation function; The influence weight of forward known data on missing data is measured by the hysteresis degradation function; Construct a monotonic matrix to determine the monotonicity of the forward known data; The missing data is completed based on the monotone matrix of the forward known data and the influence weight of the forward known data on the missing data. The expression of the monotone matrix is: Among them, i is the time point where the data is lost. If the overall running data is monotonically increasing, the monotonic matrix ζ τ >0; if the overall running data is monotonically decreasing, the monotonic matrix ζ τ <0; The calculation formula for completing the missing data is: in, is the data after the missing data at time point i is completed, θ1 and θ2 are the forward known running data y i-1 and y i-2 For lost data y i Impact weight; is the global average, that is, the average of all forward known data; l i =1, indicating that the data at the i-th time point is lost.

2. The method for future blind missing data completion for real-time equipment health monitoring according to claim 1 is characterized in that: The indicator L is a binary vector used to indicate whether the data is lost. i =1 indicates the data y at the i-th time point i Lost, l i =0 means the data y at time point i i Not lost.

3. The method for future blind missing data completion for real-time equipment health monitoring according to claim 1 is characterized in that: The autocorrelation of the forward preset number of non-missing running data is read through the autocovariance function. For the running data {y t }, the autocorrelation function of lag p is expressed as: Among them, ɑ p is the autocovariance function of lag p, ɑ0 is the autocovariance function of zero lag, n is the number of running data, p represents the number of selected time lags, y t is the running data value at time t, y t-p Running data value at time tp, is the average of the running data; combined with the missing data indicator L, if and only if l t and l t-p When both are 0, η p Only then will there be calculated value.

4. The method for future blind missing data completion for real-time equipment health monitoring according to claim 1 is characterized in that: The process of fitting the weight matrix and bias vector parameter values in the hysteresis degradation function includes: In order to combine the data autocorrelation information, take θ = η p , read the data degradation information, that is, use the following model to fit the unknown parameters: Among them, ψ p is the distance metric vector corresponding to the running data under the lag p step; The loss function is constructed using the square error, and the weight matrix a and bias vector b are calculated by minimizing the loss. The specific calculation formula is: Substitute the calculated weight matrix a and bias vector b into the hysteresis degradation function.

5. A future blind lost data completion device for real-time health monitoring of equipment, characterized by: The method adopts the future blind lost data completion method for real-time health monitoring of equipment according to any one of claims 1 to 4, comprising: Data acquisition module, used to read real-time operating data of mechanical equipment; A data loss classification module, for classifying whether the real-time running data is lost using an indicator; The distance measurement module is used to construct a distance measurement matrix when missing data occurs to quantify the degree of correlation between the missing data and the forward known data; The data correlation description module is used to construct a hysteresis degradation function based on the distance measurement matrix to describe the degradation relationship between data; read the autocorrelations of the forward known data, fit the weight matrix and bias vector parameter values in the hysteresis degradation function; and measure the influence weight of the forward known data on the missing data through the hysteresis degradation function; Data monotonicity determination module, used to construct a monotonic matrix and determine the monotonicity of forward known data; The data completion module is used to complete the missing data based on the monotone matrix of the forward known data and the influence weight of the forward known data on the missing data.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the future blind lost data completion method for real-time health monitoring of equipment as described in any one of claims 1 to 4 are implemented.

Images