Health index construction method for residual life prediction of major equipment

Abstract

The invention discloses a health index construction method for residual life prediction of major equipment, and particularly relates to the field of life prediction. Comprising the following steps: analyzing correlation among a plurality of sensor indexes in equipment to obtain a correlation coefficient, and establishing a correlation network of the sensor indexes of the equipment according to the correlation coefficient; in the correlation network, determining a dynamic model of each sensor index according to influence values of other sensor indexes on the sensor index; determining a steady-state value of each sensor index according to the kinetic model; determining an equipment state value according to the steady-state values of all the sensor indexes; according to the equipment state value, determining an influence value of each sensor index on the equipment state, and according to the influence value, determining a fusion coefficient of each sensor index; and carrying out weighted fusion on the fusion coefficients of all sensor indexes to obtain the health indexes of the equipment, so that the health state change of the equipment can be better reflected.

Description

Technical Field

[0001] This application relates to the field of life prediction, and particularly to a method for constructing health indicators for the remaining life prediction of major equipment. Background Art

[0002] Fault prediction and health management of major equipment have become an important means to improve the reliability, safety and economy of equipment. The rapid development of sensor technology, data acquisition technology, data analysis technology and artificial intelligence technology provides strong technical support for fault prediction and health management of major equipment. These technologies make it possible to monitor the operation status of equipment in real time, predict potential faults and take corresponding maintenance measures. As one of the keys to fault prediction and health management, remaining life prediction aims to predict the time when equipment may fail and its remaining life. Accurate remaining life prediction can greatly support maintenance decisions, facilitate taking preventive maintenance measures in advance, and reduce maintenance costs, thus effectively avoiding safety problems and economic losses caused by equipment failures. Therefore, developing effective remaining life prediction methods is crucial for ensuring the safe operation of major equipment.

[0003] The construction of health indicators is one of the important tasks for remaining life prediction. By constructing accurate and reliable health indicators, the current state of equipment can be evaluated more effectively, thus providing strong support for remaining life prediction. When monitoring the health status of equipment, it usually includes multi-source sensor data. The sensor monitoring data itself contains the changes in the health status of equipment. Single-sensor data often partially reflects the potential changes in the health status of equipment. At the same time, there are mutual influences among sensor data. One sensor indicator may positively or negatively affect another sensor indicator. When constructing existing health indicators, the influences among sensor data are not considered, and the obtained health indicators cannot accurately reflect the changes in the health status of equipment. Summary of the Invention

[0004] The main purpose of this application is to provide a method for constructing health indicators for the remaining life prediction of major equipment, aiming to solve the problem that the health indicators constructed by existing methods cannot accurately reflect the changes in the health status of equipment.

[0005] To achieve the above object, the present application provides a method for constructing health indicators for predicting the remaining life of major equipment, including: analyzing the correlation between multiple sensor indicators in the equipment to obtain correlation coefficients, and establishing a correlation network of the equipment sensor indicators according to the correlation coefficients; in the correlation network, determining the dynamic model of each sensor indicator according to the influence value of other sensor indicators on this sensor indicator; determining the steady-state value of each sensor indicator according to the dynamic model; determining the equipment state value according to the steady-state values of all the sensor indicators; determining the influence value of each sensor indicator on the equipment state according to the equipment state value, and determining the fusion coefficient of each sensor indicator according to the influence value; performing weighted fusion on the fusion coefficients of all the sensor indicators to obtain the health indicator of the equipment.

[0006] Optionally, determining the equipment state value according to the steady-state values of all the sensor indicators includes: determining the equipment state value according to the steady-state values of all the sensor indicators, the weighted degree of each sensor indicator in the correlation network, and a first preset relationship.

[0007] Optionally, the first preset relationship is:

[0008]

[0009] where

[0010] In the formula, k i is the weighted degree of the sensor indicator i in the correlation network, is the steady-state value of the sensor indicator i, <k>is the average weighted degree, N is the number of sensor metrics, and x eff is the equipment status value.

[0011] Optionally, the kinetic model is:

[0012]

[0013] In the formula, represents the positive correlation between sensor metrics i and j, represents the negative correlation between sensor metrics i and j, and B i is the reaction rate, N is the number of sensor metrics, and x i (t) is the value of sensor metric i at monitoring time t.

[0014] Optionally, according to the equipment status value and the influence value of each sensor metric on the equipment status, determine the fusion coefficient of each sensor metric, including: removing each sensor metric in the correlation network one by one, determining the equipment status value corresponding to the removal of each sensor metric, using the difference between the equipment status value and the equipment status value corresponding to the removal of each sensor metric as the influence value of each sensor metric on the equipment status, and determining the fusion coefficient of each sensor metric according to the influence value of each sensor metric on the equipment status.

[0015] Optionally, the calculation formula for the fusion coefficient of each sensor metric is:

[0016]

[0017] In the formula, sgn(·) is the sign function, x eff is the equipment status value, is the equipment status value after removing node i, and x i (t) is the value of sensor metric i at monitoring time t.

[0018] Optionally, analyze the correlation between multiple sensor metrics, including: calculating the correlation between each sensor metric using the Pearson correlation coefficient to obtain the correlation network of the equipment sensor metrics.

[0019] Optionally, before analyzing the correlation between multiple sensor metrics in the equipment, the method further includes: normalizing and smoothing filtering each sensor metric in turn.

[0020] Compared with the prior art, the beneficial effects of this application are as follows:

[0021] The method for constructing a health index for predicting the remaining life of major equipment according to the present invention uses a network dynamics model to describe the coupling mechanism between multi-source data, realizes the fusion of multi-source data by calculating the fusion coefficient, and constructs a composite health index that can reflect the performance degradation of the equipment; it is applicable to characterizing the health state of equipment with multi-source sensor monitoring, and there is no need to extract specific indexes that can reflect the change of its health state according to the research object, and it has good generalization; the interaction between sensor indexes is described by a dynamics model, and a multi-dimensional performance index network formed by the interaction of multi-sensor indexes is established, which can better obtain the influence degree of a single sensor index on the overall performance of the system. Taking the influence values of each sensor index as the fusion coefficient for the fusion of multi-source data, a composite performance index that can more accurately reflect the change of the equipment health state can be obtained, and further the accuracy of predicting the remaining life of the equipment can be improved. Description of the Drawings

[0022] Figure 1 It is a schematic flowchart of a method for constructing a health index for predicting the remaining life of major equipment according to the present application;

[0023] Figure 2 It is a comparison chart of the original sensor data and the processed sensor data in Embodiment 1 of the present application;

[0024] Figure 3 It is a correlation result chart between each sensor index in Embodiment 1 of the present application;

[0025] Figure 4 It is a correlation network diagram obtained in Embodiment 1 of the present application;

[0026] Figure 5 It is a bar chart of the fusion coefficient in Embodiment 1 of the present application;

[0027] Figure 6 It is a schematic diagram of the health index in Embodiment 1 of the present application;

[0028] Figure 7 It is a comparison chart of the health index constructed in Embodiment 1 of the present application and the performance of each sensor index;

[0029] Figure 8 It is a comparison chart of the health index constructed in Embodiment 1 of the present application and the performance of other health indexes;

[0030] Figure 9 It is a comparison chart of the health index constructed in Embodiment 1 of the present application and the life prediction effect of each sensor index;

[0031] Figure 10 It is a comparison chart of the health index constructed in Embodiment 1 of the present application and the life prediction effect of other health indexes.

[0032] The realization, functional features, and advantages of the present application will be further described in conjunction with embodiments and with reference to the accompanying drawings. Specific Embodiments

[0033] To make the objectives, technical solutions, and advantages of the present application clearer, the technical solutions in the present application will be clearly and completely described below in conjunction with the accompanying drawings in the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts belong to the scope of protection of the present application.

[0034] The first embodiment of the present invention provides a method for constructing health indicators for predicting the remaining life of major equipment, as Figure 1 shown, which specifically includes the following steps:

[0035] Step S1: Analyze the correlation between multiple sensor indicators of the equipment to obtain a correlation coefficient, and establish a correlation network of the equipment sensor indicators based on the correlation coefficient;

[0036] Step S11: Normalize and smooth-filter multiple sensor indicators;

[0037] In this embodiment, since the operation of major equipment involves multiple parameters and states, different sensor data often have different dimensions and orders of magnitude. Directly processing data with different dimensions and orders of magnitude easily causes the data of some sensors to be overwhelmed by the data of other sensors. Therefore, it is necessary to normalize the original data. This embodiment uses the maximum-minimum normalization method to scale the data to the [0, 1] scale. In addition, due to the influence of various interferences and noises, multi-source sensing data often has large randomness and volatility. It is necessary to further perform data smoothing and filtering on the monitoring data of each sensor of each randomly degraded equipment to reduce the influence of data randomness and volatility on data modeling and prediction. This embodiment uses the moving average method for processing. After processing, the data that originally had different scales and large fluctuations is scaled to the same scale and becomes smoother, facilitating subsequent operations.

[0038] Step S12: Analyze the correlation between the processed multiple sensor indicators to obtain a correlation coefficient, and establish a correlation network of the equipment sensor indicators based on the correlation coefficient;

[0039] Specifically, different sensors may measure different physical quantities or parameters, and there may be correlations between these parameters. This embodiment uses the Pearson correlation coefficient to calculate the correlation between each sensor indicator to obtain a correlation coefficient. The Pearson correlation coefficient calculation formula is:

[0040]

[0041] wherein represents the time series data of sensor i, T is the end time, i = 1, …, N, and N is the number of sensor indicators. If there is data of multiple pieces of equipment of the same type, then x i (t) is the mean value of the data of the same sensor indicator of all equipment at the same monitoring time. ρ(X i , X j ) describes the degree of correlation between sensor indicators i and j.

[0042] Based on ρ(X i , X j ), a weighted adjacency matrix A for constructing an index correlation network can be obtained, and the element A ij is expressed as follows:

[0043]

[0044] wherein, A ij describes the relationship between indicators i and j, represents positive correlation, represents negative correlation. Regarding the corresponding sensor indicators as nodes, if A ij ≠0, it indicates that there is a link between them. Based on the weighted adjacency matrix A, a correlation network of equipment sensor indicators can be obtained.

[0045] Step S2, in the correlation network, determine the dynamic model of each sensor indicator according to the influence value of other sensor indicators on this sensor indicator; further characterize the coupling between sensor indicators through network dynamics. Specifically, the dynamic model is:

[0046]

[0047] wherein represents the positive correlation between sensor indicators i and j, represents the negative correlation between sensor indicators i and j, B i is the reaction rate, N is the number of sensor indicators, and x i (t) is the sensor indicator i at the monitoring time t. In order to describe the coupling between each sensor indicator using the above dynamic model, it is necessary to calculate B i for each sensor indicator. When the system reaches a steady state, dx i (t) / dt = 0, and thus B i Value. The second term in the above kinetic model represents the influence of the indicators that are positively correlated with indicator i on it, and the third term represents the influence of the indicators that are negatively correlated with indicator i on it.

[0048] Step S3: Determine the steady-state value of each sensor indicator according to the kinetic model, and determine the equipment status value according to the steady-state values of all sensor indicators;

[0049] Specifically, determine the equipment status value according to the steady-state value of each sensor indicator, its weighted degree in the correlation network, and the first preset relationship. The first preset relationship is:

[0050]

[0051] where

[0052] In the formula, k i is the weighted degree of sensor indicator i in the correlation network, is the steady-state value of sensor indicator i, <k>is the average weighted degree, N is the number of sensor indicators, and x eff is the equipment status value.

[0053] It should be noted that the steady-state value of the sensor indicator in this embodiment is the numerical value of the sensor indicator corresponding to the equipment reaching the steady state described by the dynamic model.

[0054] Step S4: Determine the influence value of each sensor indicator on the equipment status according to the equipment status value, and determine the fusion coefficient of each sensor indicator according to the influence value;

[0055] Specifically, all the sensor indicators in the correlation network are removed one by one, and the equipment status value corresponding to the removal of each sensor indicator is determined. The difference between the equipment status value and the equipment status value corresponding to the removal of each sensor indicator is used as the influence value of each sensor indicator on the equipment status, and the fusion coefficient of each sensor indicator is determined according to the influence value of each sensor indicator on the equipment status.

[0056] Furthermore, the calculation formula for the fusion coefficient of each sensor indicator is:

[0057]

[0058] In the formula, sgn(·) is the sign function. If the corresponding sensor data is monotonically decreasing, its value is -1; otherwise, it is 1. x eff is the equipment status value, is the equipment status value after removing node i, x i (t) is the value of sensor indicator i at monitoring time t.

[0059] Step S5: Perform weighted fusion on the fusion coefficients of all sensor indicators to obtain the health indicator of the equipment, reducing the N-dimensional sensor data to one dimension. The calculation of the health indicator is:

[0060]

[0061] In the formula, w i is the health indicator.

[0062] Using the health indicator obtained in this embodiment to predict the remaining life of major equipment, the prediction method specifically includes the following steps:

[0063] Step S6: Use the Wiener process to model the evolution process of the health indicator over time. The model is as follows:

[0064] Z m (t) = z m,0 + λ m t + σ m B(t)

[0065] where \(z\) m,0 is the initial degradation of equipment \(m\) at \(t\) m,0 = 0, \(\lambda\) m is the drift coefficient, characterizing the degradation rate of the equipment, and \(\sigma\) m is the diffusion coefficient, describing the stochastic uncertainty of the degradation process, and \(B(t)\) is the standard Brownian motion describing the time-varying stochasticity of the degradation process.

[0066] Step S7, based on the concept of first passage time, the equipment life is defined as: \(Y\) m = inf\(\{t:Z\) m (t)\(\geq v|z\) m,0 < \(v\}\), where \(v\) is the failure threshold. The life \(Y\) m follows an inverse Gaussian distribution, and its mathematical expectation and variance are:

[0067]

[0068] where \(v\), \(\lambda\) m , \(\sigma\) m are unknown variables, and \(\lambda\) m , \(\sigma\) m can be obtained through maximum likelihood estimation as follows:

[0069]

[0070] where \(\Delta t\) j = \(t\) m,j - \(t\) m,j-1 , \(j = 1,\ldots,K\) m , \(K\) m + 1 = \(T\) is the length of the time series of the health index of equipment \(m\), that is, the number of monitoring data.

[0071] Step S8, according to the predicted life value of equipment \(m\) can be expressed as follows:

[0072]

[0073] To solve for \(v\), the following objective function is established:

[0074]

[0075] where \(M\) is the number of equipment, and \(y\) m is the actual life value. The above formula characterizes the error between the predicted life value and the actual value. Using the quasi-Newton method to solve the minimization of \(J(v)\) to obtain the optimal solution \(v\) * .

[0076] Step S9, the health index, the failure threshold \(v\) * , substitute into the formula for the predicted value of the life, and obtain the predicted value of the life of the equipment.

[0077] It should be noted that the existing method of constructing health indicators and predicting life through deep learning is similar to a black box model. It uses algorithms to extract features, making it difficult to intuitively understand the relationship between the input and output. Moreover, deep learning algorithms involve a large number of parameters that need to be continuously debugged and optimized. However, due to its complex mechanism of action, it is difficult to track the impact of each parameter on the final result. Compared with this method, the method of constructing health indicators in this embodiment can clearly describe the calculation process of each sensor indicator fusion coefficient and has interpretability.

[0078] Case analysis: C-MAPSS turbofan engine dataset

[0079] In this embodiment, the C-MAPSS turbofan engine public dataset is used as an example to construct health indicators and predict the life of the equipment. This embodiment uses the method of statistical modeling for life prediction, which has natural advantages in dealing with degradation characteristics with complete monitoring data and certain trends to achieve parametric modeling and quantify prediction uncertainty. If the engine stops too early, the degradation trend may not appear. Therefore, the full life cycle data of 100 engines from the start of service to failure in the FD001 dataset are mainly used, including the monitoring data of 21 sensors. They are divided into a training set and a test set according to the ratio of 6:4. The detailed descriptions of the 21 sensor indicators are shown in Table 1. In the original data, some sensor data do not have obvious time-varying characteristics, such as sensors 1, 5, 6, 10, 16, 18, 19. Therefore, this part of sensor data is not considered in the subsequent analysis. The remaining original data and processed data are as Figure 2 shown.

[0080] Table 1 Sensor indicator description

[0081]

[0082]

[0083] First, preprocess the Figure 2 data in a. Scale the sensor data with different dimensions and orders of magnitude to the [0, 1] scale through max-min normalization. In addition, use the method of moving average for noise reduction to reduce the randomness and volatility of the original data. The obtained sensor indicators are shown in Figure 2 a.

[0084] Step S10, calculate the correlation between each sensor indicator after preprocessing, and the obtained correlation coefficient results are as Figure 3 shown. Figure 3 Taking the T24 sensor metrics as an example, the correlation between it and other sensor metrics is shown. As can be seen from the figure, there is a positive correlation between it and some sensors, and a negative correlation between it and some other sensors. A correlation network is established based on the correlation coefficients, as Figure 4 shown. Among them, the nodes represent sensor metrics, the edges represent the relationships existing between sensor metrics, red represents negative correlation, and black represents positive correlation.

[0085] Step S20: In the correlation network, according to the influence value of other sensor metrics on this sensor metric, determine the dynamic model of each sensor metric, and characterize the coupling between sensor metrics through the dynamic model;

[0086] Step S30: Determine the steady-state value of each sensor metric according to the dynamic model, and determine the equipment state value according to the steady-state values of all sensor metrics;

[0087] Step S40: Determine the fusion coefficient of each sensor metric according to the equipment state value and the influence value of each sensor metric on the equipment state under overall coupling. The result is as Figure 5 shown.

[0088] Step S50: According to the calculated fusion coefficients, fuse the multi-source sensor data to obtain a one-dimensional composite health metric, as Figure 6 shown.

[0089] Step S60: Evaluate the performance of the health metric; common evaluation methods include monotonicity, time correlation, consistency, and mixed evaluation metrics, as follows.

[0090] Health metric performance evaluation metric 1: Monotonicity, calculated as follows:

[0091]

[0092] In the formula, and respectively represent the number of dZ m >0 and dZ m <0, dZ m is the difference in the health metric of the equipment at adjacent time points, and T is the length of the time series of the corresponding health metric, that is, the number of monitoring data. The larger the value of Mon m , the better the monotonicity.

[0093] Health metric performance evaluation metric 2: Time correlation, calculated as follows:

[0094]

[0095] In the formula, and respectively represent the sorted sequences of the health index of equipment m and the operation time l. Corr m The larger the value of Corr, the stronger the correlation between the health index and the operation time.

[0096] Health index performance evaluation index 3: Consistency, which is calculated as follows:

[0097]

[0098] In the formula, P EoL and P0 respectively represent the vectors composed of the health indexes of different equipment at the end of the service life and the initial moment, std(·) represents the standard deviation, and mean(·) represents the mean value. The larger the value of Con, the better the consistency of the health index.

[0099] Health index performance evaluation index 4: Hybrid evaluation index, which is calculated as follows:

[0100]

[0101] In the formula, M is the number of equipment, α i >0, let α1 = α2 = α3 = 1 / 3. Mon m and Con are calculated by formulas (11) and (13) respectively. Mon′ m is calculated by the following formula:

[0102]

[0103] In the formula, d 2 Z m is the second-order difference of the health index. The hybrid index evaluates the comprehensive performance of the health index from multiple aspects. The larger the value of Hyb, the better the performance of the health index.

[0104] Compare the composite health index obtained in this embodiment with the results of the health indexes constructed by the single sensor index, the data fusion method based on the genetic algorithm (Comparison 1), the semi-parametric data fusion method (Comparison 2), the data fusion method based on grey relational analysis (Comparison 3), and the variable selection and data fusion method based on relative entropy (Comparison 4). Figure 7 is the result of comparison with the single sensor index. Figure 7 In the abscissa of, HI refers to the health index constructed in this embodiment, and the rest are the indexes of each sensor. In the legend, Mon, Corr, Con, and Hyb are the average monotonicity, average time correlation, consistency, and hybrid evaluation index respectively. From Figure 7 As can be seen from the results, the values of the health indicators constructed in this embodiment are greater than the corresponding values of each sensor indicator in terms of average monotonicity, average time correlation, consistency, and mixed evaluation indicators. Therefore, the performance of the constructed health indicators is better than that of single-sensor indicators. Figure 8 are the results of comparing with the health indicators constructed by other methods. From Figure 8 the results, it can be seen that the health indicators proposed in this invention are better than other compared health indicators.

[0105] Based on the constructed health indicators, the degradation process of the system is modeled using the Wiener process, and the quasi-Newton method is used to optimize and solve the objective function on the training set to obtain the optimal failure threshold v * . Then, according to the obtained failure threshold, the life of the equipment is predicted on the test set using the prediction formula, and the following evaluation indicators are used to evaluate the life prediction effect.

[0106] Life prediction effect evaluation indicator 1: Root mean square error, and the calculation formula is as follows:

[0107]

[0108] Life prediction effect evaluation indicator 2: Mean absolute error, and the calculation formula is as follows:

[0109]

[0110] Life prediction effect evaluation indicator 3: Mean absolute percentage error, and the calculation formula is as follows:

[0111]

[0112] The smaller the values of the above evaluation indicators, the better the life prediction effect. Figure 9 shows the comparison of the life prediction effects of the health indicators constructed in this embodiment and the single-sensor indicators. Figure 9 In the abscissa of, HI refers to the health indicator constructed in this invention, and the rest are each sensor indicator. From the results, it can be seen that the constructed health indicator has a better effect than the single-sensor indicator. Figure 10 shows the comparison of the life prediction effects of the health indicators constructed in this invention and other health indicators. From the results, it can be seen that the health indicator constructed in this invention has a better effect.

[0113] The above is only the preferred embodiment of this application, and it does not limit the patent scope of this application. Any equivalent structure or equivalent process transformation made using the content of the specification and drawings of this application, or directly or indirectly applied in other related technical fields, is equally included in the patent protection scope of this application.< / k> < / k>

Claims

1. A method for constructing health indicators for predicting the remaining life of major equipment, characterized in that Including: Analyze the correlation between multiple sensor metrics in the equipment to obtain a correlation coefficient, and establish a correlation network of the equipment sensor metrics according to the correlation coefficient; In the correlation network, determine the dynamic model of each sensor metric according to the influence value of other sensor metrics on this sensor metric; Determine the steady-state value of each sensor metric according to the dynamic model; Determine the equipment status value according to the steady-state values of all sensor metrics; Determine the influence value of each sensor metric on the equipment status according to the equipment status value, and determine the fusion coefficient of each sensor metric according to the influence value; Perform weighted fusion on the fusion coefficients of all sensor metrics to obtain the health metric of the equipment.

2. The method for constructing health indicators for predicting the remaining life of major equipment according to claim 1, wherein The determining the equipment status value according to the steady-state values of all sensor metrics includes: Determine the equipment status value according to the steady-state values of all sensor metrics, the weighted degree of each sensor metric in the correlation network, and a first preset relationship.

3. The method for constructing health indicators for remaining life prediction of major equipment according to claim 2, characterized in that, The first preset relationship is: Among them, where k i is the weighted degree of sensor index i in the correlation network, is the steady-state value of sensor index i, <k>is the average weighted degree, N is the number of sensor indicators, and x eff is the equipment status value.< / k> 4. The method for constructing health indicators for remaining life prediction of major equipment according to claim 1, characterized in that The dynamic model is: In the formula, represents the positive correlation between sensor indicators i and j, represents the negative correlation between sensor indicators i and j, B i is the reaction rate, N is the number of sensor indicators, x i (t) is the value of sensor indicator i at the monitoring time t.

5. The method for constructing health indicators for predicting the remaining life of major equipment according to claim 1, characterized in that, The determining the influence value of each sensor metric on the equipment status according to the equipment status value, and determining the fusion coefficient of each sensor metric according to the influence value includes: Remove each sensor metric in the correlation network one by one, and determine the equipment status value corresponding to the removal of each sensor metric; Take the difference between the equipment status value and the equipment status value corresponding to the removal of each sensor metric as the influence value of each sensor metric on the equipment status; Determine the fusion coefficient of each sensor metric according to the influence value.

6. The method for constructing health indicators for predicting the remaining life of major equipment according to claim 5, characterized in that The calculation formula for the fusion coefficient of each sensor metric is: where sgn(·) is the sign function, and x eff is the equipment status value, is the equipment status value after removing node i, and x i (t) is the value of sensor index i at the monitoring time t.

7. The method for constructing health indicators for predicting the remaining life of major equipment according to claim 1, characterized in that, The analyzing the correlation between multiple sensor metrics includes: Calculate the correlation between each sensor metric by using the Pearson correlation coefficient to obtain the correlation network of the equipment sensor metrics.

8. The method for constructing health indicators for predicting the remaining life of major equipment according to claim 1, wherein Before the analyzing the correlation between multiple sensor metrics in the equipment, the method further includes: Perform normalization and smoothing filtering processing on each sensor metric in turn.

Images