Rotary machinery health index construction method based on reference conversion neural network under time-varying rotating speed working condition

Abstract

The invention relates to a rotating machine health index construction method based on a benchmark conversion neural network under a time-varying rotating speed working condition, and belongs to the technical field of rotating machine health state evaluation. The method comprises the following steps: S1, data acquisition; s2, the collected full life cycle vibration signals are preprocessed; s3, determining a baseline rotating speed band; s4, establishing a performance degradation model under the baseline rotating speed zone; S5, after degradation model parameters under the baseline rotating speed zone are obtained, performing interpolation on a non-reference observation value according to the established double-index degradation model so as to obtain a reference state observation sequence approximate to the whole life cycle, namely a reference health index; s6, constructing, training and optimizing a benchmark conversion neural network; and S8, constructing health indexes: inputting test data into the in-service mechanical equipment, and obtaining the health indexes of the rotating machinery by using the optimized benchmark conversion neural network. The accuracy of predicting the residual service life of the rotating machine can be improved.

Description

Technical Field

[0001] The invention belongs to the technical field of rotating machinery health status assessment, and relates to a method for constructing a rotating machinery health index based on a reference conversion neural network under time-varying speed conditions. Background Art

[0002] Rotating machinery is widely used in production and life, and gears, bearings and other components are important components of its transmission chain. Due to the harsh operating environment such as speed change, heavy load, high temperature, corrosion, etc., key components such as gears and bearings of rotating machinery are prone to defects such as pitting and fracture. The expansion of these defects will destroy the smooth operation of the equipment, increase the noise and vibration amplitude, reduce the accuracy and reliability of the equipment, and eventually lead to equipment failure and even casualties. Therefore, it is very important to quantitatively evaluate the health status of rotating machinery during operation. During normal operation, mechanical equipment is not allowed to observe its health status and detect the degree of degradation in the form of frequent shutdowns. As an important bridge connecting mechanical health monitoring and remaining life prediction, health indicators can not only quantitatively reveal the health status of equipment operation, but also estimate its remaining service life with the help of specific prediction methods.

[0003] However, the existing health index construction methods are usually only applicable to mechanical equipment running at a constant speed. In actual engineering, rotating machinery often needs to change its operating conditions due to the requirements of production activities, such as the change of car gears, the change of industrial robot joint movement, and the change of wind turbine pitch. This causes the collected state monitoring signal to be interfered by the amplitude modulation of the speed change, resulting in the inability of the existing health index construction method to clearly and accurately reveal the health degradation process of the equipment, thereby affecting the judgment of the remaining life. If all state observations can be unified to a set baseline speed, the interference introduced by the speed change can be eliminated. Based on this, it is very important to study a method for constructing mechanical health indicators under time-varying speed conditions. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a method for constructing health indicators of rotating machinery based on a reference conversion neural network under time-varying speed conditions, so as to improve the accuracy of the remaining service life prediction of rotating machinery.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A method for constructing a rotating machinery health index based on a reference conversion neural network under time-varying speed conditions specifically comprises the following steps:

[0007] S1: Data collection: Use the condition monitoring system to collect the vibration signal of the mechanical equipment throughout its life cycle X = (x1, x2, ..., x N ) T, where N is the number of samples in the whole life cycle, M is the number of sampling points for a single sample; at the same time, the real-time speed of the acquisition device S = (s1, s2, ..., s N ) T And other status information;

[0008] S2: Data preprocessing: preprocess the collected full life cycle vibration signals;

[0009] S3: determine the baseline speed band;

[0010] S4: Establishing a performance degradation model under the baseline speed band: After determining the baseline speed band, given an observation value, it can be determined whether it is a baseline observation value based on the speed information; for observation values ​​in non-baseline speed bands, a performance degradation model is established to interpolate to approximate its baseline observation;

[0011] S5: Interpolation of non-reference observations: After obtaining the degradation model parameters under the baseline speed band, the non-reference observations are interpolated according to the established double exponential degradation model to obtain the reference state observation sequence that approximates the entire life cycle. That is, baseline health indicators;

[0012] S6: constructing a benchmark conversion neural network for converting the original state observation variables into virtual benchmark observation values ​​under the baseline speed;

[0013] S7: training optimized benchmark conversion neural network;

[0014] S8: Construct health indicators: For in-service mechanical equipment, input test data and use the optimized benchmark conversion neural network to obtain the rotating machinery health indicators.

[0015] Further, in step S2, data preprocessing specifically includes: filtering and denoising the collected full life cycle vibration signal, removing abnormal values ​​and speed matching, etc.; in order to reduce the data dimension, extracting the RMS value or other trend statistical features of the vibration signal as the state observation variable F = (f1, f2, ..., f N ) T .

[0016] Further, in step S3, determining the baseline speed band specifically includes: defining a speed band floating around the baseline speed:

[0017] s∈[(1-δ)×s B ,(1+δ)×s B ]

[0018] Among them, s B Represents the rated speed of the equipment, and δ is the floating coefficient. The floating coefficient δ can be set based on engineering experience or data analysis results.

[0019] Further, step S4 specifically includes: selecting a double exponential model as a degradation fitting model of the state value under the baseline speed band, and the expression is:

[0020] Y(t)=α1exp(β1t)+α2exp(β2t)+ε

[0021] Among them, Y(t) is the approximate value of the reference state observation at time t, α1, α2, β1 and β2 are state parameters; is the Gaussian noise term.

[0022] Further, in step S5, after obtaining enough state observation variables, according to the benchmark observation sequence {f *} Estimate the parameters of the degradation model:

[0023]

[0024] Among them, f i * Indicates t i The state observation value obtained at the baseline speed band at all times, represents the estimated value of the state parameter.

[0025] Furthermore, in step S6, the constructed reference conversion neural network includes an input layer, an output layer and N hidden layers; wherein the input layer has three neurons, and only three parameters need to be input during training and testing, namely, the original state observation variable f, the baseline speed ratio R s and degradation time ratio R t ; The predicted value of the output layer is the virtual benchmark observation value converted to the baseline speed band;

[0026] Baseline speed ratio R s is the current speed of the device s(t) relative to the baseline speed s B The ratio of the actual speed of the device to the specified baseline speed; the degradation time ratio R t Refers to the current running time of the device t k The ratio of the equipment to its full life cycle T contains the equipment operation time information; during training and testing, R t The calculation method is different, namely:

[0027]

[0028] Where n is the number of samples of similar equipment with known life cycle, T j is the full life cycle of the jth sample of similar equipment.

[0029] Further, in step S7, training the optimized benchmark conversion neural network specifically includes: assuming θ BTNN Represents the weights and bias parameters in the benchmark conversion neural network. During the network training process, the network's predicted value Benchmark observations that approximate the entire life cycle The mean square error between the two is taken as the optimization target:

[0030]

[0031] Among them, n batch is the batch size of the input data during training, is the objective function; the training strategy uses a small batch stochastic gradient descent algorithm to optimize the network parameters. Using the back propagation algorithm, the optimization parameters are calculated by minimizing the objective function:

[0032]

[0033] Among them, lr is the network learning rate, θ BTNN are network parameters, are the optimized network parameters; the above training process is iterated to obtain the optimized benchmark conversion neural network.

[0034] Further, in step S8, constructing the health index specifically includes: based on the optimized benchmark conversion neural network, for the in-service mechanical equipment, inputting the test data {f}, {R s}, {R t} to θ BTNN Real-time prediction of the health index value when it is converted to the baseline speed, and the life cycle health index vector H = [h1,h2,…,h N ] T , thereby revealing its performance degradation process and laying the foundation for subsequent fault warning and remaining life prediction.

[0035] The beneficial effect of the present invention is that the present invention can evaluate the health status of rotating machinery under time-varying speed conditions, improve the accuracy of predicting the remaining service life of rotating machinery, and thus reduce the maintenance cost of mechanical equipment.

[0036] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0038] Figure 1 It is a flow chart of a method for constructing health indicators of rotating machinery based on a reference conversion neural network according to the present invention;

[0039] Figure 2 Schematic diagram of the benchmark conversion neural network structure;

[0040] Figure 3 This is a simplified diagram of the turbofan engine structure;

[0041] Figure 4 To test the health indicators of the engine;

[0042] Figure 5 This is a simplified diagram of the wind turbine gearbox structure;

[0043] Figure 6 Comparison of methods for constructing different health indicators for two test bearings. DETAILED DESCRIPTION

[0044] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0045] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0046] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0047] See also Figure 1 to Figure 6 The present invention provides a method for constructing a health index of a rotating machinery based on a reference conversion neural network under time-varying speed, such as Figure 1 As shown, the specific process is:

[0048] 1) Data collection: Use the condition monitoring system to collect the vibration signal X=(x1, x2,…, x N ) T , where N is the number of samples in the whole life cycle, M is the number of sampling points for a single sample. At the same time, the real-time speed of the acquisition device S = (s1, s2, ..., s N ) T And other status information.

[0049] 2) Data preprocessing: The collected full life cycle vibration signal is preprocessed by filtering and noise reduction, removing abnormal values ​​and speed matching. In order to reduce the data dimension, the RMS value or other trend statistical features of the vibration signal are extracted as the state observation variable F = (f1, f2, ..., f N ) T .

[0050] 3) Determine the baseline speed band: Considering that rotating machinery needs to operate under rated conditions most of the time, the rated speed is specified as the baseline speed. In addition, due to factors such as transmission error and measurement error, it is not enough to just use a constant speed value as a reference benchmark, but to define a speed band that floats around the baseline speed:

[0051] s∈[(1-δ)×s B ,(1+δ)×s B ]

[0052] Among them, s B Represents the rated speed of the equipment, and δ is the floating coefficient. The floating coefficient δ can be set based on engineering experience or data analysis results.

[0053] 4) Establish a performance degradation model under the baseline speed band: After determining the baseline speed band, given an observation value, it can be determined whether it is a baseline observation value based on the speed information. For observations in non-baseline speed bands, a performance degradation model needs to be established for interpolation to approximate its baseline observation. Affected by multiple internal and external random factors, the degradation trend of rotating machinery has two common degradation modes: gradual degradation and sudden failure. The double exponential model has the advantage of expressing these two degradation modes at the same time, so it is selected as the degradation fitting model for state values ​​under the baseline speed band:

[0054] Y(t)=α1exp(β1t)+α2exp(β2t)+ε

[0055] Among them, α1, α2, β1 and β2 are state parameters; is a Gaussian noise term. After obtaining enough state observation variables, according to the benchmark observation sequence {f *} Estimate the parameters of the degradation model:

[0056]

[0057] Among them, f i * Indicates t i The state observations obtained at the baseline speed at all times.

[0058] 5) Non-benchmark observation interpolation: After obtaining the degradation model parameters under the baseline speed band, the non-benchmark observation values ​​are interpolated according to the established double exponential degradation model to obtain an approximate baseline state observation sequence for the entire life cycle. That is, the baseline health indicator.

[0059] 6) Constructing a benchmark conversion neural network: There are some problems in the actual implementation of using a benchmark conversion function to realize the mapping from the original observation to the benchmark state value. First, under the premise that the transformation factors λ(·) and μ(·) satisfy two basic restrictions, a variety of benchmark conversion functions can be constructed, and different conversion functions have different characteristics, resulting in differences in the scope of application. Secondly, with the increase of operating time, the performance of rotating machinery gradually degrades due to the expansion of damage size, and the changes in speed and load make the operating conditions worse, resulting in aggravated degradation. The benchmark conversion function may be difficult to model the accelerated degradation process caused by changes in operating conditions, and cannot reveal the complex nonlinear mapping relationship from the original state variables to the virtual benchmark observations.

[0060] Based on the above considerations, due to the powerful nonlinear approximation ability of deep neural networks, a benchmark conversion neural network is constructed to implicitly model the complex nonlinear mapping relationship between the original state variables and the benchmark observations. The network structure of the benchmark conversion neural network is as follows: Figure 2 As shown in Figure 1, it is mainly composed of an input layer, an output layer and N hidden layers. The input layer of the benchmark conversion neural network has three neurons. During training and testing, only three parameters need to be input, namely, the original state observation variable f, the baseline speed ratio R s and degradation time ratio R t The predicted value of the network output layer is the virtual baseline observation value converted to the baseline speed.

[0061] Baseline speed ratio R s is the current speed of the device s(t) relative to the baseline speed s B The ratio of the actual speed of the device to the specified baseline speed contains information about the degradation time ratio R t Refers to the current running time of the device tk The ratio of the equipment to its full life cycle T contains the equipment operation time information. It is worth noting that during the training and testing process, R t The calculation method is different, namely:

[0062]

[0063] Where n is the number of samples of similar equipment whose entire life cycle is known.

[0064] 7) Model training: Assume θ BTNN Represents the weights and bias parameters in the benchmark conversion neural network. During the network training process, the network's predicted value With the benchmark observation y * The mean square error between the two is taken as the optimization target:

[0065]

[0066] Among them, n batch The training strategy uses a small batch stochastic gradient descent algorithm to optimize network parameters. Using the back propagation algorithm, the optimization parameters can be calculated by minimizing the objective function:

[0067]

[0068] Among them, lr is the network learning rate. By iterating the above training process, an optimized benchmark conversion neural network can be obtained.

[0069] 8) Health index construction: Based on the optimized benchmark conversion neural network, the in-service mechanical equipment is input with test data {f}, {R s}, {R t} to θ BTNN Real-time prediction of the health index value when it is converted to the baseline speed, and the life cycle health index vector H = [h1,h2,…,h N ] T , thereby revealing its performance degradation process and laying the foundation for subsequent fault warning and remaining life prediction.

[0070] The above is the method for constructing the health index of rotating machinery based on the reference conversion neural network under the time-varying speed condition proposed by the present invention. The effectiveness of the method is illustrated by specific experiments below.

[0071] Embodiment 1:

[0072] like Figure 3As shown in the figure, NASA Ames Research Center simulated the damage propagation path inside the aircraft turbofan engine through the C-MAPSS simulation model, and generated 21 sensor variables to reflect the engine degradation. The C-MAPSS data set consists of four subsets FD001-FD004 generated under different combinations of six operating conditions and two failure modes. Since subsets FD001 and FD003 are single conditions, and subset FD004 is not only affected by the operating speed, but also has two failure modes, but the failure modes are not within the scope of the study. Therefore, the FD002 subset under the influence of time-varying speed is selected as verification. In order to improve computational efficiency, 5 engine unit data are randomly selected to form the data set, of which 3 groups are used for training and 2 groups are used for testing.

[0073] The turbofan engine health indicators finally constructed are as follows: Figure 4 As shown in the figure. It can be seen from the figure that compared with the original state variable, the health index constructed based on the proposed method significantly reduces the irregular fluctuations caused by the change in engine speed, presents a monotonically increasing trend, and can effectively reflect the degradation trend of the engine health status over time. In the initial stage, the number of working cycles of the turbofan engine is small and the health status is good. With the increase of running time, the value of the health index gradually increases, indicating that with the increase of working time, the health status of the turbofan engine gradually deteriorates. Finally, the value of the health index reaches the maximum value, which indicates that the turbofan engine cannot work normally due to some serious faults.

[0074] Comparative test:

[0075] In order to fully demonstrate the superiority of this method, the health index (BTNN-HI) constructed by the proposed method is compared with the original observed variables and the health index based on the benchmark transfer function (BTF-HI).

[0076] Among them, in terms of the construction of the benchmark conversion function, the present invention selects an exponential type expression and a linear expression to construct the scale transformation factor and the position transformation factor of the benchmark conversion function:

[0077] y(t)=γexp(s(t)-s B )y * (t)+ρ(s(t)-s B )

[0078] Among them, γ and ρ are parameters that need to be fitted according to known data. On the contrary, after determining the specific form of the benchmark conversion function, the state observation value at any speed can be mapped to the benchmark state value through the following inverse function:

[0079]

[0080] Among them, y kand represent the original observations and the observations converted to the baseline speed, respectively.

[0081] It should be noted that the benchmark observation can only be observed at the baseline speed, and the benchmark observation obtained at other speeds is a virtual value. For virtual observations that cannot be directly obtained, the benchmark observation is approximated by interpolating the degradation model at the baseline speed. Therefore, the parameters γ and ρ can be estimated by minimizing the sum of squares of the error between the interpolated approximate benchmark observation and the converted virtual benchmark value:

[0082]

[0083] Furthermore, the optimized estimated parameters can be solved by using the Levenberg-Marquardt algorithm or the trust region reflective algorithm, thereby obtaining the desired benchmark transfer function.

[0084] In order to comprehensively evaluate the performance of health indicators, the above three health indicators are compared from the perspectives of monotonicity (Mon), trend (Tre), robustness (Rob) and comprehensive evaluation index (CI).

[0085] In order to calculate these evaluation indicators, the health indicators need to be decomposed into trend terms and random terms:

[0086] H(t k )=H T (t k )+H R (t k )

[0087] Among them, H(t k ) indicates that at t k The health index value at the moment, H T (t k ) is its average trend, H R (t k ) is a random fluctuation. Based on the obtained trend term and random term, the calculation formulas of monotonicity, trend, robustness and comprehensive index are as follows:

[0088] 1) Monotonicity:

[0089]

[0090] 2) Trend:

[0091]

[0092] 3) Robustness:

[0093]

[0094] Among them, δ(·) is the sgn(·) function, are the order columns of H and T respectively, N is the number of health indicators, and the parameter ξ is the relaxation factor that controls the strictness of robustness.

[0095] 4) Comprehensive evaluation indicators:

[0096] CI=eta1*Mon3(H)+eta2*Tre3(H,T)+eta3*Rob2(H)

[0097] Among them, η i is the trade-off coefficient and ∑ i η i =1, monotonicity is generally considered to be more important, so the trade-off coefficients are taken as 0.4, 0.3, and 0.3 respectively.

[0098] Table 1 Performance comparison results of different health indicators of test engines E4 and E5

[0099]

[0100] The performance evaluation results of different health indicators of the two test turbofan engines are shown in Table 1. It can be seen from Table 1 that the mechanical health indicator construction method based on the benchmark conversion neural network proposed in the present invention has greatly improved its comprehensive performance compared with the original observation and the health indicator construction method based on the benchmark conversion function, and its performance is more superior. However, the performance of BTF-HI on different test engines is different, and it is not necessarily better than the original observation variables. Therefore, the proposed method is more suitable for the task of constructing health indicators of rotating machinery under time-varying speed.

[0101] Embodiment 2:

[0102] The gearbox structure of a 2MW wind turbine in a domestic wind farm is as follows: Figure 5 As shown in Figure 2, its transmission system is mainly composed of a planetary stage and two-stage parallel gear transmission. An accelerometer is installed on the bearing housing near the low-speed shaft bearing to monitor the vibration of the low-speed shaft bearing during the operation of the wind turbine. The sampling frequency of the acceleration sensor is 25.6kHz, and the length of each sampling time is 1.28s. In addition, a speed sensor is installed on the main shaft to obtain the input speed information. Since the wind turbine gearbox bearing (WTGB) data set comes from the actual engineering environment, the original vibration signal contains a lot of measurement noise and interference from other components, so filtering and noise reduction are performed. In addition, preprocessing such as abnormal data removal and speed data matching is required, and finally 5 groups of full-life data sets are obtained, as shown in Table 2.

[0103] Table 2 Detailed information of WTGB dataset

[0104]

[0105] Since the data in this embodiment are actually collected vibration signals, in addition to the original observed RMS value and BTF-HI, dimensionless health indicators such as skewness, kurtosis, and shape factor are also extracted as a comparison method. The health indicators of the test bearings B4 and B5 constructed by different methods are as follows: Figure 6 As shown. Figure 6 It can be observed that BTNN-HI, RMS and BTF-HI can better reveal the relationship between the deterioration of bearing health status over time compared with indicators such as skewness and kurtosis waveform factor. Although BTF-HI can reduce the interference of time-varying speed to a certain extent, it may also introduce new jump values. BTNN-HI can not only largely eliminate the influence of speed changes, but also has a clearer degradation path and a more consistent failure threshold at the end of life, which is particularly important for improving the accuracy of mechanical health status assessment and RUL prediction.

[0106] Table 3 Performance comparison results of different health indicators of bearing B4

[0107]

[0108] Table 4 Performance comparison results of different health indicators of bearing B5

[0109]

[0110] The quantitative evaluation results of the inherent performance of different health indicators of the two tested wind turbine bearings are shown in Tables 3 and 4, respectively. It can be found from the table that the health indicator constructed by the proposed method has the largest improvement in monotonicity compared with the other five indicators, which is also consistent with the results of intuitive observation of the health indicator curve. In addition, the trend and robustness have also been greatly improved, especially for bearing B4. Although skewness, kurtosis and waveform factor are dimensionless indicators, their performance evaluation results are relatively poor and cannot clearly characterize the degradation state of wind turbine bearings. In summary, BTNN-HI has better comprehensive performance than other health indicators, and is more conducive to mechanical health status monitoring and remaining service life prediction under time-varying speed.

[0111] From the above two applications, it can be seen that compared with health indicators such as RMS, BTF-HI, skewness, kurtosis and form factor, BTNN-HI can more clearly reveal the degradation trend of rotating machinery under time-varying speed, and it is easier to determine a unified failure threshold, thereby reducing the experimental cost. Therefore, compared with the health indicator construction method based on the benchmark transfer function and dimensionless, the benchmark transfer neural network is more suitable for constructing the health indicator of rotating machinery under time-varying speed conditions.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A method for constructing a health index of rotating machinery based on a reference conversion neural network under time-varying speed conditions, characterized in that: The method specifically comprises the following steps: S1: Data collection: Use the condition monitoring system to collect the vibration signal of the mechanical equipment throughout its life cycle X = (x1, x2, ..., x N ) T , where N is the number of samples in the whole life cycle, M is the number of sampling points for a single sample; at the same time, the real-time speed of the acquisition device S = (s1, s2, ..., s N ) T ; S2: Data preprocessing: preprocess the collected full life cycle vibration signals; S3: determine the baseline speed band; S4: Establishing a performance degradation model under the baseline speed band: After determining the baseline speed band, given an observation value, determine whether it is a baseline observation value based on the speed information; for observation values ​​in non-baseline speed bands, establish a performance degradation model to interpolate to approximate its baseline observation; S5: Interpolation of non-reference observations: After obtaining the degradation model parameters under the baseline speed band, the non-reference observations are interpolated according to the established double exponential degradation model to obtain the reference state observation sequence that approximates the entire life cycle. That is, baseline health indicators; S6: constructing a benchmark conversion neural network for converting the original state observation variables into virtual benchmark observation values ​​under the baseline speed; S7: training optimized benchmark conversion neural network; S8: Construct health indicators: For in-service mechanical equipment, input test data and use the optimized benchmark conversion neural network to obtain the rotating machinery health indicators.

2. The method for constructing a health index of a rotating machinery according to claim 1, characterized in that: In step S2, data preprocessing specifically includes: filtering and denoising the collected full life cycle vibration signal, removing abnormal values ​​and speed matching; in order to reduce the data dimension, extracting the RMS value or other trend statistical features of the vibration signal as the state observation variable F = (f1, f2, ..., f N ) T .

3. The method for constructing a health index of a rotating machinery according to claim 1, characterized in that: In step S3, determining the baseline speed band specifically includes: defining a speed band floating around the baseline speed: s∈[(1-δ)×s B ,(1+δ)×s B ] Among them, s B Represents the rated speed of the equipment, and δ is the floating coefficient.

4. The method for constructing a health index of a rotating machinery according to claim 2, characterized in that: Step S4 specifically includes: selecting a double exponential model as a degradation model of the state value under the baseline speed band, and the expression is: Y(t)=α1exp(β1t)+α2exp(β2t)+ε Among them, Y(t) is the approximate value of the reference state observation at time t, α1, α2, β1 and β2 are state parameters; is the Gaussian noise term.

5. The method for constructing a health index of a rotating machinery according to claim 4, characterized in that: In step S5, after obtaining enough state observation variables, according to the baseline observation sequence {f * } Estimate the parameters of the degradation model: Among them, f i * Indicates t i The state observation value obtained at the baseline speed band at all times, represents the estimated value of the state parameter.

6. The method for constructing a health index of a rotating machinery according to claim 1, characterized in that: In step S6, the constructed reference conversion neural network includes an input layer, an output layer and N hidden layers; wherein the input layer has three neurons, and only three parameters need to be input during training and testing, namely, the original state observation variable f, the baseline speed ratio R s and degradation time ratio R t ; The predicted value of the output layer is the virtual benchmark observation value converted to the baseline speed band; Baseline speed ratio R s is the current speed of the device s(t) relative to the baseline speed s B The ratio of the actual speed of the device to the specified baseline speed; the degradation time ratio R t Refers to the current running time of the device t k The ratio of the equipment to its full life cycle T contains the equipment operation time information; during training and testing, R t The calculation method is different, namely: Where n is the number of samples of similar equipment with known life cycle, T j is the full life cycle of the jth sample of similar equipment.

7. The method for constructing a health index of a rotating machinery according to claim 6, characterized in that: In step S7, training the optimized benchmark conversion neural network specifically includes: assuming θ BTNN Represents the weights and bias parameters in the benchmark conversion neural network. During the network training process, the network's predicted value The baseline state observation value f that is close to the full life cycle B * The mean square error between the two is taken as the optimization target: Among them, n batch is the batch size of input data during training, is the objective function; the training strategy uses a small batch stochastic gradient descent algorithm to optimize the network parameters. Using the back propagation algorithm, the optimization parameters are calculated by minimizing the objective function: Among them, lr is the network learning rate, θ BTNN are network parameters, are the optimized network parameters; the above training process is iterated to obtain the optimized benchmark conversion neural network.

8. The method for constructing a health index of a rotating machinery according to claim 7, characterized in that: In step S8, constructing the health index specifically includes: based on the optimized benchmark conversion neural network, for the in-service mechanical equipment, inputting the test data {f}, {Rs}, {Rt} to θ BTNN Real-time prediction of the health index value when it is converted to the baseline speed, and the life cycle health index vector H = [h1,h2,…,h N ] T .

Images