Equipment life prediction method based on multilevel analysis and sequential evaluation method

Abstract

The invention relates to an equipment life prediction method based on a multilevel analysis and sequential evaluation method. The invention relates to the technical field of complex equipment life prediction. The method comprises the following steps: acquiring multi-dimensional operation data of equipment and historical monitoring data of similar equipment through a sensor; performing time-frequency domain feature extraction on the acquired data, extracting data features, and storing the extracted data features; obtaining a health degree curve HD of each module of the equipment by using a sequential evaluation method; obtaining a weight coefficient weight by using a multi-level evaluation method; and calculating the health degree HI of the equipment according to the weighted fusion. The reliability and safety of the equipment are improved, the maintenance plan is optimized, the service life of the equipment is prolonged, and the fault risk is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of complex equipment life prediction, and is a method for predicting the life of equipment based on a multi-level analysis and sequential evaluation method. Background Art

[0002] In the existing methods for predicting the life of equipment indirectly by constructing an equipment health curve, the technology based on deep learning directly constructs an equipment health degree curve through a neural network. However, this method has strong data dependence, poor interpretability, and a high cost of obtaining full-cycle degradation data. In addition, the method based on linear weighted fusion is intuitive and simple to implement, but has poor anti-noise ability and is easily affected by data outliers, resulting in a decrease in the accuracy of the prediction results. Summary of the Invention

[0003] Aiming at the deficiencies of the existing technology, the present invention realizes the timely warning of the equipment failure state by constructing a health index (HI) curve of the equipment. This method can be widely applied to industries such as industry, aviation, aerospace, and transportation, aiming to improve the reliability and safety of equipment, optimize maintenance plans, extend the service life of equipment, reduce the risk of failures. The present invention provides a method for predicting the life of equipment based on a multi-level analysis and sequential evaluation method.

[0004] The present invention provides the following technical solutions:

[0005] A method for predicting the life of equipment based on a multi-level analysis and sequential evaluation method, the method comprising the following steps:

[0006] Step 1: Obtain multi-dimensional operation data of the equipment and historical monitoring data of similar equipment through sensors;

[0007] Step 2: Extract time-frequency domain features of the obtained data, extract the data features, and store them;

[0008] Step 3: Use the sequential evaluation method to obtain the health degree curve HD of each module of the equipment;

[0009] Step 4: Use the multi-level evaluation method to obtain the weight coefficient weight;

[0010] Step 5: Calculate the health index HI of the equipment according to weighted fusion.

[0011] Preferably, the specific content of step 2 is:

[0012] Extract time-frequency domain features of the obtained data, extract the data features, and store them; the determined health state features of the equipment's sequential operation data include the maximum value, minimum value, mean value, standard deviation, skewness, kurtosis, root mean square, spectral energy, and spectral entropy.

[0013] Preferably, step 3 is specifically as follows:

[0014] Step 3.1: Obtain the standard health feature vector μ U , and calculate the mean μ based on the features extracted from the health status of the device to obtain the standard health feature vector:

[0015] μ U = [μ mean , μ std , μ skew , μ kurtosis , μ max , μ min , μ rms , μ spectralenergy , μ spectralentropy

[0016] Step 3.2: Obtain the device health memory matrix ∑U, obtain the features obtained from the health data of the initial period of the device, calculate its covariance matrix, and highlight the correlation between different features;

[0017] Step 3.3: Perform Mahalanobis distance calculation to obtain the Mahalanobis distance between the feature vector of the device at the current moment and the standard health feature vector, which is used to measure the device state deviation. The Mahalanobis distance is calculated by the following formula:

[0018]

[0019] where x is the feature vector of the device at the current moment;

[0020] Step 3.4: Calculate the ratio SPR of sequential evaluation, and calculate the wallpaper of the Mahalanobis distance of the device state at the current moment and the Mahalanobis distance of the initial state of the device to obtain the sequential evaluation ratio, which is used to describe the damage state of the device:

[0021]

[0022] Step 3.5: Calculate the health degree HD of the device module. The size of the health degree describes the health status of the device:

[0023]

[0024] where α is a tension parameter.

[0025] Preferably, the health degree HD of the device module: above 0.8 represents the healthy state, and below 0.4 represents the warning state.

[0026] Preferably, step 4 is specifically as follows:

[0027] Step 4.1: Construct a hierarchical structure, including robustness, predictability, trendiness, monotonicity, and temporal correlation;​

[0028] Step 4.2: Construct the comparison matrix A for each index. By means of pairwise comparison, construct the judgment matrix between each module, and use the 1-4 scale method for comparison. The larger the value, the more important it is. The calibrated value is determined according to the size of the calculated index to obtain A Rob , A Pre , A Tre , A Mon and A Corr , all of which have dimensions of N×N, where N represents the number of modules of the device;

[0029] Step 4.3: Construct the index importance comparison matrix B to describe the sensitivity of the device to each index;

[0030] Step 4.4: Perform weight calculation according to the comparison matrix A of each index and the index importance comparison matrix B; Step 4.5: Perform consistency test. The consistency test of the matrix ensures the correctness of the pairwise comparison process:

[0031] The consistency index CI is calculated as follows:

[0032]

[0033] where λ max is the maximum eigenvalue of the comparison matrix, and n is the dimension of the matrix;

[0034] The consistency ratio CR is calculated as follows:

[0035]

[0036] where RI is the random consistency index, which depends on the dimension of the comparison matrix. When CR≤0.1, the consistency passes; otherwise, adjust the comparison matrix;

[0037] Step 4.6: Obtain the comprehensive weight coefficient weight, expressed as [weight 1 , weight 2 , …, weight N , where N represents the number of modules of the device, and is calculated by the following formula:

[0038] weight = weight Rob × w B1 + weight Pre × w B2 + weight Tre × w B3 + weight Mon × w B4 + weight Corr × w B5 .

[0039] Preferably, the robustness is calculated by the following formula:

[0040]

[0041] where x i is the health data point of the device module with noise, μ is the mean health of the device module with noise, μ no-noise is the mean health of the device module without noise, and N is the size of the data window;

[0042] The predictability index is calculated by the following formula:

[0043]

[0044] where represents the monitoring data characteristics at the device failure moment, represents the monitoring data characteristics at the initial startup moment of the device;

[0045] The trendiness is calculated by the following formula:

[0046]

[0047] The monotonicity is calculated by the following formula:

[0048]

[0049] where X(t j ) represents the monitoring data of the device at time t j , and sgn() represents the sign function.

[0050] Preferably, according to the comparison matrix A of each index and the index importance comparison matrix B, the corresponding weight matrices are obtained respectively, which are represented as weight Rob , weight Pre , weight Tre , weight Mon , weight Corr and weight B , the weight dimension size corresponding to the index is 1×N, and the weight dimension size corresponding to the index importance is 1×5. The calculation uses the eigenvector method. By performing eigenvalue decomposition on the comparison matrix, the eigenvector is obtained and normalized to obtain the weight size, where weight B is represented as [w B1 , w B2 , w B3 , w B4 , w B5 .

[0051] An equipment life prediction system based on a multi-level analysis and sequential evaluation method, the system comprising:

[0052] A data monitoring module, which acquires multi-dimensional operation data of the equipment and historical monitoring data of similar equipment through sensors;

[0053] A feature extraction module, which extracts time-frequency domain features from the acquired data, extracts data features, and stores them;

[0054] A health curve acquisition module, which uses a sequential evaluation method to acquire the health degree curve HD of each module of the equipment;

[0055] A weight module, which uses a multi-level evaluation method to acquire a weight coefficient weight;

[0056] A weighted fusion module, which calculates the health degree HI of the equipment according to weighted fusion.

[0057] A computer-readable storage medium, on which a computer program is stored, and the program is executed by a processor to implement an equipment life prediction method based on a multi-level analysis and sequential evaluation method.

[0058] A computer device, comprising a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it implements an equipment life prediction method based on a multi-level analysis and sequential evaluation method.

[0059] The present invention has the following beneficial effects:

[0060] Compared with the prior art, the present invention:

[0061] The present invention extracts data features in the time-frequency domain. Secondly, it uses a sequential evaluation method to acquire the health curves of each module of the equipment. Then, according to the data features and the health curves of the modules, it dynamically acquires the weight coefficients of each module by using a multi-level analysis method. Finally, it performs weighted fusion to obtain the final equipment health degree curve. There are mainly two innovation points in this method. Compared with the deep learning method for the health curve acquisition method, it does not rely on a large amount of data. The multi-level analysis method comprehensively utilizes five criteria of robustness, predictability, trendiness, monotonicity, and time series correlation, has stronger interpretability, obtains the equipment health degree curve more reasonably, gives an early warning in the equipment failure state, and predicts the equipment life. Description of the Drawings

[0062] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0063] Figure 1 It is shown as the method flowchart of the present invention;

[0064] Figure 2 It is shown as the effect comparison diagram of the present invention.

[0065] Figure 3 It is shown as the result diagram of the equipment health index construction of the present invention;

[0066] Figure 4 It is shown as the result diagram of the equipment health index construction of the present invention. Specific Embodiments

[0067] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the drawings. Obviously, the described embodiments are some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0068] The following describes the present invention in detail in combination with specific embodiments. Specific Embodiment 1:

[0070] According to Figures 1 to 4 As shown, the specific optimization technical solution adopted by the present invention to solve the above technical problems is: The present invention relates to a method for predicting the equipment life based on a multi-level analysis and sequential evaluation method.

[0071] The present invention provides a method for predicting the equipment life based on a multi-level analysis and sequential evaluation method, and the method includes the following steps:

[0072] Step 1: Obtain the multi-dimensional operation data of the equipment and the historical monitoring data of the same type of equipment through sensors;

[0073] Step 2: Extract the time-frequency domain features of the obtained data, extract the data features, and store them;

[0074] Step 3: Use the sequential evaluation method to obtain the health degree curve HD of each module of the equipment;

[0075] Step 4: Use the multi-level evaluation method to obtain the weight coefficient weight;

[0076] Step 5: Calculate the health index HI of the device based on weighted fusion. Specific Embodiment 2:

[0078] The difference between Embodiment 2 and Embodiment 1 of this application is only that:

[0079] The specific content of Step 2 is:

[0080] Extract the time-frequency domain features of the acquired data, store the extracted data features; the determined health status features of the device's time-series operation data include maximum value, minimum value, mean value, standard deviation, skewness, kurtosis, root mean square, spectral energy, and spectral entropy. Specific Embodiment 3:

[0082] The difference between Embodiment 3 and Embodiment 2 of this application is only that:

[0083] The specific content of Step 3 is:

[0084] Step 3.1: Obtain the standard health feature vector μ U , calculate the mean μ according to the features extracted from the health status of the device, and obtain the standard health feature vector:

[0085] μ U = [μ mean , μ std , μ skew , μ kurtosis , μ max , μ min , μ rms , μ spectralenergy , μ spectralentropy

[0086] Step 3.2: Obtain the device health memory matrix ∑U, obtain the features obtained from the device's initial period health data, calculate its covariance matrix, and highlight the correlation between different features;

[0087] Step 3.3: Perform Mahalanobis distance calculation, obtain the Mahalanobis distance between the device's current moment feature vector and the standard health feature vector, and use it to measure the device state deviation. The Mahalanobis distance is calculated by the following formula:

[0088]

[0089] Among them, x is the feature vector of the device at the current moment;

[0090] Step 3.4: Calculate the ratio SPR of sequential evaluation, calculate the ratio of the Mahalanobis distance of the device state at the current moment to the Mahalanobis distance of the device's initial state, and obtain the sequential evaluation ratio to describe the damage state of the device:

[0091]

[0092] Step 3.5: Calculate the health degree HD of the device module. The magnitude of the health degree describes the health status of the device:

[0093]

[0094] where α is a tension parameter. Specific Embodiment Four:

[0096] The difference between Embodiment Four and Embodiment Three of this application lies only in:

[0097] For the health degree HD of the device module: above 0.8 represents a healthy state, and below 0.4 represents a warning state. Specific Embodiment Five:

[0099] The difference between Embodiment Five and Embodiment Four of this invention lies only in:

[0100] Specifically, Step 4 is as follows:

[0101] Step 4.1: Construct a hierarchical structure, including robustness, predictability, trendiness, monotonicity, and temporal correlation;

[0102] Step 4.2: Construct a comparison matrix A for each index. By means of pairwise comparison, construct a judgment matrix between each module, and use the 1-4 scale method for comparison. The larger the value, the more important it is. The calibrated value is determined according to the magnitude of the calculated index to obtain A Rob 、A Pre 、A Tre 、A Mon and A Corr , all of which have dimensions of N×N, where N represents the number of modules of the device;

[0103] Step 4.3: Construct an index importance comparison matrix B to describe the sensitivity of the device to each index;

[0104] Step 4.4: Perform weight calculation according to the comparison matrix A of each index and the index importance comparison matrix B; Step 4.5: Perform consistency check. The consistency check of the matrix ensures the correctness of the pairwise comparison process:

[0105] The consistency index CI is calculated as follows:

[0106]

[0107] where λ max is the maximum eigenvalue of the comparison matrix, and n is the dimension of the matrix;

[0108] The consistency ratio CR is calculated as follows:

[0109]

[0110] Among them, RI is the random consistency index, which depends on the dimension of the comparison matrix. When CR ≤ 0.1, the consistency passes; otherwise, the comparison matrix is adjusted.

[0111] Step 4.6: Obtain the comprehensive weight coefficient weight, expressed as [weight 1 , weight 2 , …, weight N , where N represents the number of modules of the device and is calculated by the following formula:

[0112] weight = weight Rob × w B1 + weight Pre × w B2 + weight Tre × w B3 + weight Mon × w B4 + weight Corr × w B5 . Specific Example Six:

[0114] The difference between Example Six and Example Five of the present invention lies only in:

[0115] The robustness is calculated by the following formula:

[0116]

[0117] Among them, x i is the health data point of the device module with noise, μ is the mean health of the device module with noise, and μ no-noise is the mean health of the device module without noise, and N is the size of the data window;

[0118] The predictability index is calculated by the following formula:

[0119]

[0120] Among them, represents the monitoring data characteristics at the device failure moment, represents the monitoring data characteristics at the initial startup moment of the device;

[0121] The trendiness is calculated by the following formula:

[0122]

[0123] The monotonicity is calculated by the following formula:

[0124]

[0125] Among them, X(t j ) represents the monitoring data of device t j at time t, and sgn() represents the sign function. Specific Embodiment Seven:

[0127] The difference between the seventh embodiment and the sixth embodiment of the present invention lies only in:

[0128] According to the comparison matrix A of each index and the comparison matrix B of index importance, the corresponding weight matrices are obtained respectively, and are represented as weight Rob , weight Pre , weight Tre , weight Mon , weight Corr and weight B . The size of the weight dimension corresponding to the index is 1×N, and the size of the weight dimension corresponding to the index importance is 1×5. The calculation uses the eigenvector method. By performing eigenvalue decomposition on the comparison matrix, the eigenvector is obtained and standardized to obtain the weight size, where weight B is represented as [w B1 , w B2 , w B3 , w B4 , w B5 . Specific Embodiment Eight:

[0130] The difference between the eighth embodiment and the seventh embodiment of the present invention lies only in:

[0131] The present invention provides a device life prediction system based on a multi-level analysis and sequential evaluation method, and the system includes:

[0132] A data monitoring module, which obtains multi-dimensional operation data of the device and historical monitoring data of similar devices through sensors;

[0133] A feature extraction module, which extracts time-frequency domain features from the acquired data, extracts the data features, and stores them;

[0134] A health curve acquisition module, which obtains the health curve HD of each module of the device by using the sequential evaluation method;

[0135] A weight module, which obtains the weight coefficient weight by using the multi-level evaluation method;

[0136] A weighted fusion module, which calculates the health index HI of the device according to weighted fusion. Specific Embodiment Nine:

[0138] The only difference between Embodiment Nine and Embodiment Eight of the present invention lies in:

[0139] The present invention provides a computer-readable storage medium, on which a computer program is stored, and the program is executed by a processor to implement a device life prediction method based on a multi-level analysis and sequential evaluation method. Specific Embodiment Ten:

[0141] The only difference between Embodiment Ten and Embodiment Nine of the present invention lies in:

[0142] The present invention provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, a device life prediction method based on a multi-level analysis and sequential evaluation method is implemented. Specific Embodiment Eleven:

[0144] The only difference between Embodiment Eleven and Embodiment Ten of the present invention lies in:

[0145] The device life prediction method proposed by the present invention can generally be divided into four steps. First, extract the data characteristics in the time-frequency domain. Secondly, use the sequential evaluation method to obtain the health curves of each module of the device. Then, based on the data characteristics and the health curves of the modules, use the multi-level analysis method to dynamically obtain the weight coefficients of each module. Finally, perform weighted fusion to obtain the final device health degree curve. The main innovations of this method are mainly two. Compared with the deep learning method for obtaining the health curve, it does not rely on a large amount of data. The multi-level analysis method comprehensively uses five criteria: robustness, predictability, trendiness, monotonicity, and time series correlation, has stronger interpretability, obtains the device health degree curve more reasonably, gives an early warning in the device failure state, and predicts the device life.

[0146] Step 1: Use sensors to obtain the multi-dimensional operation data of the device and the historical monitoring of similar devices.

[0147] Step 2: Extract the time-frequency domain features of the obtained data, and store the extracted data features. The determined health state features of the device's time series operation data include the maximum value (max), minimum value (min), mean value (mean), standard deviation (std), skewness, kurtosis, root mean square (RMS), spectral energy, and spectral entropy.

[0148] Step 3: Use the sequential evaluation method to obtain the health degree curves HD of each module of the device. The specific steps are as follows:

[0149] (1) Obtain the standard health feature vector μ U . Calculate the mean μ based on the features extracted from the health status of the device to obtain the standard health feature vector.

[0150] μ U = [μ mean , μ std , μ skew , μ kurtosis , μ max , μ min , μ rms , μ spectralenergy , μ spectralentropy

[0151] (2) Obtain the device health memory matrix ΣU. Obtain the features from the health data of the initial period of the device, calculate its covariance matrix, and highlight the correlation between different features.

[0152] (3) Perform Mahalanobis distance calculation to obtain the Mahalanobis distance between the feature vector of the device at the current moment and the standard health feature vector, which is used to measure the device state deviation. The Mahalanobis distance calculation formula is as follows:

[0153]

[0154] where x is the feature vector of the device at the current moment.

[0155] (4) Calculate the ratio SPR of sequential evaluation. Calculate the ratio of the Mahalanobis distance of the device state at the current moment and the Mahalanobis distance of the initial state of the device to obtain the sequential evaluation ratio, which is used to describe the damage state of the device.

[0156]

[0157] (5) Calculate the health degree HD of the device module. The size of the health degree describes the health status of the device. Generally, above 0.8 represents a healthy state, and below 0.4 represents a warning state.

[0158]

[0159] where α is a tension parameter used to control the influence of the sequential evaluation ratio on the health degree, which can be determined according to expert knowledge or the device data monitoring interval.

[0160] Step 4: Obtain the weight coefficient weight using the multi-level assessment method (MAPH). The specific steps are as follows:

[0161] (1) Construct the hierarchical structure.

[0162] ​In this method, the target layer of the hierarchical analysis is to obtain the weight coefficients of each module of the device. The criterion layer consists of various indicators for health assessment, including five indicators: robustness (Rob), predictability (Pre), trendiness (Tre), monotonicity (Mon), and temporal correlation (corr). The scheme layer includes the characteristics of each module of the device and the health curve.

[0163] (2) Construct the comparison matrix A for each indicator, which describes the importance of device modules under a specified indicator. By using the pairwise comparison method, construct the judgment matrix between each module, and use the 1-4 scale method for comparison. The larger the value, the more important it is. The calibrated value is determined according to the size of the calculated indicator to obtain A Rob 、A Pre 、A Tre 、A Mon and A Corr , all of which have dimensions of N×N, where N represents the number of modules of the device.

[0164] Robustness (Rob) reflects the stability of the health of device modules in the face of data noise and outliers. The calculation formula is as follows:

[0165]

[0166] where x i is the health data point of the device module with noise, μ is the mean health of the device module with noise, μ no-noise is the mean health of the device module without noise, and N is the size of the data window.

[0167] Predictability (Pre) refers to the dispersion of data and the range of data variability at the moment of device failure. The predictability indicator is calculated as follows:

[0168]

[0169] where, represents the monitoring data characteristics at the moment of device failure, represents the monitoring data characteristics at the initial startup moment of the device.

[0170] Trendiness (Tre) refers to the change trend of device health over time. By fitting the change trend of device health, calculate the slope of health over time, where t represents time.

[0171]

[0172] Monotonicity (Mon) describes the overall degradation trend of device performance and the smoothness of data.

[0173]

[0174] Among them, X(t j ) represents the monitoring data of device t j at time t, and sgn() represents the sign function.

[0175] The temporal correlation (corr) measures the correlation between the module health and time using the Pearson correlation coefficient.

[0176] (3) Construct the index importance comparison matrix B, which describes the sensitivity of the device to each index and can be constructed based on expert knowledge.

[0177] (4) Conduct weight calculation. According to the comparison matrix A of each index and the index importance comparison matrix B, the corresponding weight matrices are obtained respectively, denoted as weight Rob , weight Pre , weight Tre , weight Mon , weight Corr and weight B . The weight dimension size corresponding to the index is 1×N, and the weight dimension size corresponding to the index importance is 1×5. The calculation uses the eigenvector method. By performing eigenvalue decomposition on the comparison matrix, the eigenvector is obtained and normalized to obtain the weight size. Among them, weight B is expressed as [w B1 , w B2 , w B3 , w B4 , w B5 .

[0178] (5) Conduct consistency test. Judging the consistency test of the matrix ensures the correctness of the pairwise comparison process.

[0179] The consistency index CI is calculated as follows:

[0180]

[0181] Among them, λ max is the maximum eigenvalue of the comparison matrix, and n is the dimension of the matrix.

[0182] The consistency ratio CR is calculated as follows:

[0183]

[0184] Among them, RI is the random consistency index, which depends on the dimension of the comparison matrix. When CR ≤ 0.1, the consistency passes; otherwise, the comparison matrix is adjusted.

[0185] (6) Obtain the comprehensive weight coefficient weight, expressed as [weight 1 , weight 2 , …, weight N , where N represents the number of modules of the device. The calculation is as follows:

[0186] weight = weight Rob × w B1 + weight Pre × w B2 + weight Tre × w B3 + weight Mon × w B4 + weight Corr × w B5

[0187] Step Five: Calculate the health index HI of the device according to the weighted fusion. Specific Example Twelve:

[0189] The difference between the twelfth embodiment of the present invention and the eleventh embodiment is only that:

[0190] A device life prediction method based on multi-level analysis and sequential evaluation method includes the following steps:

[0191] Step One: Use sensors to obtain the multi-dimensional operation data of the device and the historical monitoring of similar devices. This method is verified using FD001 of the n-campass dataset.

[0192] Step Two: Eliminate the unchanging sensor data, extract the time-frequency domain features of the remaining sensor data, and store the extracted data features. The determined health state features of the device's time-series operation data include amplitude, maximum value, minimum value, mean value, standard deviation, skewness, kurtosis, root mean square, spectral energy, and spectral entropy.

[0193] Step Three: Use the sequential evaluation method to obtain the health index curves of each sensor of the device. The tension parameter α is set to 0.08.

[0194] Step Four: Use the multi-level evaluation method to obtain the weight coefficients.

[0195] Step Five: Obtain the health index curve by weighted fusion. The obtained health index curve has good smooth monotonicity and can respond in time when the device is approaching the failure moment. Set 0.4 as the warning line, and the accuracy of timely warning is over 90%.

[0196] The above is only a preferred embodiment of a device life prediction method based on a multi-level analysis and sequential evaluation method. The protection scope of a device life prediction method based on a multi-level analysis and sequential evaluation method is not limited to the above embodiments. Any technical solutions falling within this concept belong to the protection scope of the present invention. It should be noted that for those skilled in the art, several improvements and changes made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. A method for predicting equipment life based on multi-level analysis and sequential evaluation method, characterized by: The method comprises the following steps: Step 1: Obtain multi-dimensional operation data of the equipment and historical monitoring data of similar equipment through sensors; Step 2: Extract the time-frequency domain features of the acquired data, extract the data features, and store them; Step 3: Use the sequential evaluation method to obtain the health curve HD of each module of the equipment; Step 4: Use the multi-level evaluation method to obtain the weight coefficient weight; Step 5: Calculate the health level HI of the device based on weighted fusion.

2. The method according to claim 1, characterized in that: The step 2 is specifically as follows: The acquired data is subjected to time-frequency domain feature extraction, and the extracted data features are stored; the health status features of the determined equipment timing operation data include maximum value, minimum value, mean value, standard deviation, skewness, kurtosis, root mean square, spectrum energy and spectrum entropy.

3. The method according to claim 2, characterized in that: The step 3 is specifically as follows: Step 3.1: Obtain the standard health feature vector μ U , according to the features extracted from the health status of the equipment, the mean μ is calculated to obtain the standard health feature vector: m U =[μ mean ,m std ,m skew ,m kurtosis ,m max ,m min ,m rms ,m spectralenergy ,m spectralentropy ] Step 3.2: Get the device health memory matrix∑ U , obtain the features obtained by using the health data of the equipment in the initial period, calculate its covariance matrix, and highlight the correlation between different features; Step 3.3: Perform Mahalanobis distance calculation to obtain the Mahalanobis distance between the current feature vector of the device and the standard health feature vector to measure the device state deviation. The Mahalanobis distance is calculated by the following formula: Among them, x is the feature vector of the device at the current moment; Step 3.4: Calculate the sequential evaluation ratio SPR, calculate the Mahalanobis distance of the current device state and the Mahalanobis distance of the initial state of the device, and obtain the sequential evaluation ratio to describe the damage state of the device: Step 3.5: Calculate the health HD of the device module. The size of the health HD describes the health status of the device: Among them, α is a tension parameter.

4. The method according to claim 3, characterized in that: Health HD of the equipment module: 0.8 or above indicates a healthy state, and 0.4 or below indicates a warning state.

5. The method according to claim 4, characterized in that: The step 4 is specifically as follows: Step 4.1: Construct a hierarchy including robustness, predictability, trend, monotonicity, and temporal correlation; Step 4.2: Construct the comparison matrix A of each indicator. Through the method of pairwise comparison, construct the judgment matrix between each module. Use the 1-4 scaling method for comparison. The larger the value, the more important it is. The calibration value is determined according to the size of the calculated indicator to obtain A. Rob , A Pre , A Tre , A Mon and A Corr , whose dimensions are all N×N, where N represents the number of modules of the device; Step 4.3: Construct the indicator importance comparison matrix B to describe the sensitivity of the device to each indicator; Step 4.4: Perform weight calculation based on the comparison matrix A of each indicator and the comparison matrix B of the indicator importance; Step 4.5: Perform consistency test, and the consistency test of the judgment matrix ensures the correctness of the pairwise comparison process: The consistency index CI is calculated as follows: Among them, λ max is the maximum eigenvalue of the contrast matrix, n is the dimension of the matrix; The consistency ratio CR is calculated as follows: Among them, RI is the random consistency index, which depends on the dimension of the contrast matrix. When CR≤0.1, the consistency is passed, otherwise the contrast matrix is ​​adjusted; Step 4.6: Get the comprehensive weight coefficient weight, expressed as [weight1, weight2, …, weight N ] Where N represents the number of modules of the device, calculated by the following formula: weight=weight Rob ×w B1 +weight Pre ×w B2 +weight Tre ×w B3 +weight Mon ×w B4 +weight Corr ×w B5 。 6. The method according to claim 5, characterized in that: The robustness is calculated by: Among them, x i is the health data point of the noisy equipment module, μ is the mean health value of the noisy equipment module, μ no-noise is the mean value of the health of the equipment module in the absence of noise, and N is the size of the data window; The predictability index is calculated as follows: in, Represents the monitoring data characteristics at the time of equipment failure, Indicates the monitoring data characteristics of the device at the initial startup time; The trend is calculated by the following formula: Monotonicity is calculated as follows: Among them, X(t j ) indicates device t j The monitoring data at the moment, sgn() represents the sign function.

7. The method according to claim 4, characterized in that: According to the comparison matrix A of each indicator and the comparison matrix B of the indicator importance, the corresponding weight matrices are obtained, which are expressed as weight Rob 、weight Pre 、weight Tre 、weight Mon 、weight Corr and weight B The weight dimension size corresponding to the indicator is 1×N, and the weight dimension size corresponding to the indicator importance is 1×5. The calculation adopts the eigenvector method. By performing eigenvalue decomposition on the comparison matrix, the eigenvector is obtained and normalized to obtain the weight size, where weight B It is expressed as [w B1 ,w B2 ,w B3 ,w B4 ,w B5 ].

8. An equipment life prediction system based on multi-level analysis and sequential evaluation method, characterized by: The system comprises: A data monitoring module, which obtains multi-dimensional operation data of the equipment and historical monitoring data of similar equipment through sensors; A feature extraction module, which extracts time-frequency domain features from the acquired data, extracts data features, and stores them; A health curve acquisition module, wherein the health curve acquisition module uses a sequential evaluation method to obtain a health curve HD of each module of the device; A weight module, wherein the weight module uses a multi-level evaluation method to obtain a weight coefficient weight; A weighted fusion module is used to calculate the health level HI of the device according to weighted fusion.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the method according to claims 1-7.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method of claims 1-7 is implemented.

Images