Rotor life prediction method and system based on big data

Abstract

The invention relates to the technical field of life prediction, in particular to a rotor life prediction method and system based on big data, and the method comprises the steps: collecting the operation parameters and working condition parameters of a rotor, and carrying out the clustering of the working condition parameters, and obtaining a plurality of working conditions; and for any working condition, a target feature vector at each moment is constructed, and the target feature vector comprises the operation parameters, the health state of the rotor and the health attenuation index of the motor. According to the method, comprehensive characterization of the degradation state of the rotor is realized by constructing comprehensive health characteristics fusing rotor vibration and the electrical state (three-phase current and power factor) of the motor. And meanwhile, the operation conditions are divided by using a clustering algorithm, and an exclusive prediction model is trained for each condition, so that the prediction precision and generalization ability of the model under complex and changeable conditions are greatly improved, and the life prediction is more accurate and reliable.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of life prediction. More particularly, the present application relates to a rotor life prediction method and system based on big data. BACKGROUND

[0002] The rotor is the core bearing component of various rotating machines (such as steam turbines, aero-engines, motors, etc.), and its health state is directly related to the safe and reliable operation of the entire equipment. In the long-term service process, the rotor will gradually produce damage and accumulate due to fatigue, wear, corrosion and other factors, and eventually lead to sudden failure, causing serious safety accidents and huge economic losses. Therefore, it is of great significance to monitor the health of the rotor and accurately predict its remaining useful life for realizing predictive maintenance, ensuring equipment safety and reducing operation and maintenance costs.

[0003] The prior art usually adopts machine learning algorithms such as random forest to construct a prediction model by fusing multiple sensor information. However, these methods have inherent defects in application: on the one hand, they mostly only rely on the mechanical signals such as vibration and temperature of the rotor itself, ignoring the coupling effect of the health state of the driving system (such as the motor) on the rotor load and dynamic characteristics, resulting in incomplete feature information input into the model. On the other hand, the rotor often experiences multiple changing operating conditions (such as different rotating speeds and loads) in actual operation, and the equipment degradation patterns and data characteristics under different operating conditions differ significantly. If a single model is used to uniformly process and predict the data under all operating conditions, it is difficult to capture the degradation law under a specific operating condition, which will lead to poor generalization ability of the prediction model and low accuracy. SUMMARY

[0004] The present application provides a rotor life prediction method and system based on big data, aiming to solve the problem in the related art that if a single model is used to uniformly process and predict the data under all operating conditions, it is difficult to capture the degradation law under a specific operating condition, which will lead to poor generalization ability of the prediction model and low accuracy.

[0005] In a first aspect, the present application provides a rotor life prediction method based on big data, comprising: collecting operating parameters and working condition parameters of the rotor, and clustering the working condition parameters to obtain multiple working conditions; for any working condition, constructing a target feature vector at each time, the target feature vector comprising: operating parameters, a health state of the rotor, and a health degradation index of the motor; wherein the health state of the rotor reflects the energy proportion of the power spectrum of the rotor vibration signal at the time in the health characteristic frequency band set; the method for obtaining the health characteristic frequency band set comprises: selecting a health reference time according to the motor health index, and constructing a reference power spectrum using the vibration signal thereof; then performing peak value detection in the reference power spectrum, identifying peak values greater than a threshold, and constructing narrow bands according to the characteristic frequencies of the peak values to obtain the health characteristic frequency band set for the working condition, wherein the motor health index reflects the imbalance degree of the three-phase current of the motor and the deviation degree of the power factor, and the deviation degree of the motor health index relative to the health reference value is taken as the health degradation index of the motor at the time; inputting the target feature vector at the current time into the residual useful life prediction model corresponding to the working condition to output the predicted residual useful life of the rotor. By combining the health state of the driving motor (quantified by the three-phase current imbalance degree and the power factor deviation) with the rotor vibration state, a more comprehensive target feature vector is constructed, the influence of the driving system on the rotor health is captured for the first time from the system level, and the problem of incomplete feature information is solved. In addition, the method clusters the working condition parameters and constructs a dedicated prediction model for each working condition, thereby improving the accuracy and reliability of the rotor residual useful life prediction.

[0006] Further, the method for constructing a reference power spectrum using a vibration signal comprises: performing Fourier transform on the vibration signal of each health reference time to obtain a power spectrum; and calculating the median of the power spectrum at each frequency point to obtain the reference power spectrum. By calculating the "median" of the power spectrum at each frequency point of multiple health times to construct the reference power spectrum, this method can effectively suppress accidental noise interference and transient fluctuations, thereby providing a solid foundation for improving the accuracy of rotor health state evaluation.

[0007] Further, the method for calculating the motor health index comprises: taking the product of the reciprocal of the three-phase current imbalance degree and the reciprocal of the deviation between the actual power factor of the motor and the rated power factor as the motor health index. Compared with a single evaluation index, this index can more comprehensively and accurately reflect the overall health level of the motor, thereby providing a more reliable input for subsequent construction of a system-level feature vector.

[0008] Further, the method for obtaining a health reference value comprises: obtaining the motor health index of the motor in the early running stage, and selecting the maximum motor health index as the health reference value.

[0009] Further, the method further comprises: determining a working condition to which a current time belongs, a distance between the current working condition parameter and a cluster center of all clustering clusters, and selecting a working condition corresponding to a clustering cluster with the closest distance as the working condition to which the current time belongs.

[0010] Further, the method further comprises: clustering the working condition parameters to obtain multiple working conditions, comprising: clustering the working condition parameters by using a K-Means algorithm to obtain multiple clustering clusters corresponding to the multiple working conditions, wherein the number of the clustering clusters is determined by using an elbow method.

[0011] Further, the method further comprises: performing smoothing processing on the time sequence of the motor health indexes before calculating the health degradation indexes of the motor at each time, which can effectively filter out short-term jitter caused by measurement noise or slight fluctuations in working conditions, thereby extracting a smooth signal that can better reflect the real and long-term degradation trend of the equipment, so that the calculated health degradation indexes are more monotonic and stable, and the robustness of the life prediction is improved. Further, the method further comprises: constructing narrow bands according to the characteristic frequencies of the peaks, comprising: constructing narrow bands with a width of ±1 Hz centered on each characteristic frequency.

[0012] Further, the early running stage is a stage within a preset time period after the first running of the motor.

[0013] The second aspect of the present application also provides a rotor life prediction system based on big data, comprising a processor and a memory, the memory stores a computer program, and the processor executes the computer program to realize the rotor life prediction method based on big data according to any one of the above.

[0014] Beneficial effects: by constructing a comprehensive health feature that integrates rotor vibration and motor electrical state (three-phase current, power factor), the degradation state of the rotor is comprehensively characterized. At the same time, a clustering algorithm is used to divide the running working conditions, and a dedicated prediction model is trained for each working condition, which greatly improves the prediction accuracy and generalization ability of the model under complex and variable working conditions, making the life prediction more accurate and reliable. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a flowchart that schematically shows rotor life prediction according to an embodiment of the present application; Figure 2 is a flowchart that schematically shows calculation of the health degradation index of the motor at each time according to an embodiment of the present application. DETAILED DESCRIPTION

[0016] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0017] As Figure 1As shown, S101: Collects rotor operating data and divides operating conditions.

[0018] Specifically, various sensors deployed on rotating machinery collect operating parameters and condition parameters covering the entire lifecycle of the rotor. These operating parameters include, but are not limited to, vibration signal characteristics, main bearing temperature, gearbox oil temperature, and rotor temperature; the condition parameters include rotational speed and load. To eliminate the influence of differences in dimensions and numerical ranges among different features, the collected raw data needs to be standardized. As a preferred approach, this embodiment employs the Z-score standardization method to ensure that all features are on the same comparable scale.

[0019] Next, the operating conditions of the rotor are categorized. Specifically, unsupervised clustering algorithms are used to cluster the operating parameters, resulting in multiple clusters corresponding to various operating conditions. For example, the K-Means algorithm can be used to divide historical data into K independent clusters, each cluster representing a specific operating condition. It should be noted that the number of clusters K can be determined using the elbow method.

[0020] S102: Construct the target feature vectors under various working conditions.

[0021] For each working condition identified in the above steps, a target feature vector for model training is independently constructed. Taking time t under the k-th working condition as an example, the target feature vector at time t is constructed. , , For the first The rotor temperature at time t under the given operating condition. For the first The temperature of the main bearing at time t under the given operating condition. For the first The ambient temperature of the rotor at time t under the given operating conditions. Indicates the first The health status of the rotor at time t under various operating conditions. For the first The motor health degradation index at time t under various operating conditions.

[0022] like Figure 2 As shown, S1021: Calculate the health status of the rotor at each moment.

[0023] Specifically, the rotor's health status is an indicator that quantifies the rotor's current health level. The logic behind constructing this health status is as follows: In a healthy rotor, the vibration energy is mainly concentrated at the rotational frequency and a few characteristic frequencies such as its low harmonics; however, when the rotor experiences early faults (such as microcracks or localized wear), it will trigger a nonlinear response, causing the vibration energy to diffuse into a wider frequency band or non-harmonic frequencies. Therefore, the rotor's health status can be assessed by quantifying the degree of energy concentration.

[0024] In this embodiment, the formula for calculating the rotor's health status is as follows: In the formula, For the first The health status of the rotor at time t under various operating conditions. For the first The power spectral density function of the vibration signal at time t under the given working condition. For the first The set of frequency bands corresponding to health characteristics under various operating conditions, which includes multiple narrow bands. This is the lower limit of narrowband. The upper limit of narrowband, For the first The sum of vibrational energy at time t under various operating conditions within the set of healthy characteristic frequency bands. For the first The total energy of the vibration signal at time t under a certain working condition.

[0025] As can be seen from the above formula, when the rotor is in a very healthy state, its vibration energy is highly concentrated in the healthy characteristic frequency band set. At this point, the numerator is approximately equal to the denominator. The value approaches 1. As equipment degrades, energy leaks to other frequency bands, causing molecules to shrink, thus... The value then decreases. Therefore, The degree of reduction directly reflects the degree of deterioration in the rotor's health condition.

[0026] No. The method for obtaining the set of healthy characteristic frequency bands corresponding to various operating conditions includes: Under the k-th operating condition, calculating the motor health index of the rotor at each moment, and selecting the N moments with the largest motor health index as the health reference moments, where N ranges from 50 to 100. Then, treating the motor at the health reference moments as a motor operating under normal conditions, statistically analyzing the rotational speed at each health reference moment under the k-th operating condition and calculating its corresponding fundamental frequency. Since it is the same operating condition, these fundamental frequencies are relatively similar. Performing a Fourier transform on the vibration signal at each moment to obtain the power spectrum. Then, calculating the median of these power spectra at each frequency point to construct a reference power spectrum representing the vibration characteristics of a healthy rotor under that operating condition.

[0027] Finally, peak detection is performed on the baseline power spectrum to identify all significant, stable energy peaks. For example, the threshold of peak detection can be set as 3 times the median of the whole baseline power spectrum higher than the peak value. For each detected peak, its frequency position is determined as a characteristic frequency, and a narrow band is constructed around the characteristic frequency, for example, a narrow band with a width of ±1 Hz can be constructed. All narrow bands are merged to obtain the health characteristic frequency band set under the kth working condition .

[0028] For the health index of the motor at each time, the calculation formula is: . In the formula, represents the motor health index at the tth time, 、 and respectively represent the three-phase current of the motor at the tth time, represents the average value of the three-phase current of the motor at the tth time, the actual power factor of the motor at the tth time, represents the rated power factor under the load at the tth time, is a preset parameter to avoid the denominator being 0.

[0029] In the formula, the deviation of the three-phase current from the average value is calculated, which reflects the degree of current imbalance, the greater the value of , the greater the current imbalance, and the smaller the value of

[0030] S1022: Calculate the motor health degradation index at each time.

[0031] First, the health baseline value of the motor is determined, specifically by finding the maximum value of the motor health index in the early running stage of the motor. The early running stage can be a stage within a preset period after the first running of the motor, and the preset period can be 10 days or a month, etc.

[0032] It should be noted that considering that the instantaneous motor health index may fluctuate due to small fluctuations in working conditions or measurement noise, in order to more accurately capture the long-term and stable degradation trend, the time series of the motor health index needs to be smoothed. The smoothing method can use the exponential moving average method for smoothing, and the exponential moving average is a commonly used smoothing method, which will not be described in detail here.

[0033] Then, the motor health degradation index at each time can be calculated based on the health benchmark value and the smoothed health index. The basis for its construction is that the degradation index should be the normalized deviation of the smoothed health index relative to the benchmark health value. In the initial stage of motor operation, its health state is close to optimal, and the smoothed health index at each time is closer to the health benchmark value of the motor. At this time, the value of the motor health degradation index is close to 0, indicating that almost no health degradation has occurred. With the increase of operation time, the performance of the motor gradually degrades, resulting in a downward trend of the health index, and the smoothed health index decreases accordingly, thereby causing the motor health degradation index to monotonically increase from 0 to 1. When the motor completely fails, the smoothed health index approaches 0, and the motor health degradation index at this time approaches 1.

[0034] S103: Construct a residual useful life prediction model under each working condition.

[0035] An independent residual useful life prediction model is trained for each working condition. In this embodiment, the random forest algorithm is selected as the prediction model. Random forest is an ensemble learning algorithm that improves the accuracy and robustness of the model by constructing multiple decision trees and integrating their prediction results (such as voting or averaging). It is also suitable for the high-dimensional, nonlinear features constructed in this embodiment.

[0036] First, a training data set is constructed. For each working condition, its training data set is composed of a set of label pairs {( , )}. represents the target feature vector at time t in the historical kth working condition, represents the real residual useful life of the rotor at time t, . In the formula, is the failure time or replacement time of the rotor in the historical log, is the timestamp of a certain historical time.

[0037] Then, the constructed training data set is divided into a training set and a validation set. Next, the key hyperparameters of the random forest are set, including the number of decision trees, the maximum depth of each tree, the maximum number of features considered when dividing nodes, and the minimum number of samples in leaf nodes. The training data is input into the model, and the mean squared error or mean absolute error is used as the loss function to optimize the division strategy of each tree. Finally, when the pre-set number of decision trees in the random forest have all been constructed based on the training data set, and the performance evaluation of the validation set meets the expected index requirements, it is considered that the model training is complete, and the residual useful life prediction model is obtained. At this time, the obtained residual useful life prediction model can be used to input the target feature vector at any time, and output the corresponding residual useful life prediction value.

[0038] S104: predicting the life of the current rotor based on the remaining useful life prediction model.

[0039] After the training of the dedicated model of all working conditions is completed, the system can be used for real-time remaining useful life prediction of the rotor. The specific process is as follows: first, the working condition parameters (rotational speed, load) and operation parameters at the current time are collected in real time, and the target feature vector at the current time is constructed according to the above steps. Second, compare the current working condition parameters with the K cluster centers obtained in step S101, select the cluster closest to the current working condition parameters as the working condition to which the current time belongs. Then input the target feature vector at the current time into the remaining useful life prediction model, and output the predicted remaining useful life of the rotor at the current time.

[0040] The application also provides a rotor life prediction system based on big data. The system comprises a processor and a memory, and the memory stores computer program instructions which, when executed by the processor, implement the rotor life prediction method based on big data according to the first aspect of the application.

[0041] The system also comprises a communication bus and a communication interface and other components familiar to those skilled in the art, the settings and functions of which are known in the art, and thus will not be described here.

[0042] In the present application, the aforementioned memory can be any tangible medium containing or storing a program, which can be used by or in conjunction with an instruction execution system, device or apparatus. For example, the computer readable storage medium can be any suitable magnetic storage medium or magneto-optical storage medium, such as resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (EDRAM), high bandwidth memory (HBM), hybrid memory cube (HMC) and the like, or any other medium that can be used to store the required information and can be accessed by an application, module or both. Any such computer storage medium can be part of the device or accessible or connectable to the device. Any application or module described in the present application can be implemented using computer readable / executable instructions stored or otherwise held by such computer readable medium.

[0043] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation to the scope of the application. It should be noted that for those skilled in the art, without departing from the concept of the present application, several modifications and improvements can be made, which are all within the protection scope of the present application.

Claims

1. A rotor life prediction method based on big data, characterized in that, The rotor life prediction method includes: collecting the rotor's operating parameters and working condition parameters, and clustering the working condition parameters to obtain multiple working conditions; For any operating condition, a target feature vector is constructed for each moment. The target feature vector includes: operating parameters, rotor health status, and motor health decay index. The rotor health status reflects the energy proportion of the power spectrum of the rotor vibration signal within the set of health feature frequency bands at that moment. The method for obtaining the set of health characteristic frequency bands includes: selecting a health reference time based on the motor health index and constructing a reference power spectrum using its vibration signal; then performing peak detection in the reference power spectrum to identify peaks greater than a threshold, constructing a narrow band based on the characteristic frequency of the peak, and merging them to obtain the set of health characteristic frequency bands under this working condition. The motor health index reflects the degree of imbalance of the three-phase current of the motor and the degree of deviation of the power factor, and the degree of deviation of the motor health index from its health reference value is used as the health attenuation index of the motor at that time. Input the target feature vector at the current moment into the remaining service life prediction model for the corresponding working condition, and output the predicted remaining service life of the rotor.

2. The rotor life prediction method based on big data according to claim 1, characterized in that, A reference power spectrum is constructed using its vibration signal, including: The power spectrum is obtained by performing Fourier transform on the vibration signals at each health reference time. The median of the power spectrum at each frequency point is calculated to obtain the reference power spectrum.

3. The rotor life prediction method based on big data according to claim 1, characterized in that, The calculation method for the motor health index includes: The product of the reciprocal of the three-phase current imbalance and the reciprocal of the deviation between the actual power factor and the rated power factor of the motor is used as the motor health index.

4. The rotor life prediction method based on big data according to claim 1, characterized in that, Methods for obtaining health baseline values ​​include: Obtain the motor health index during the early operation phase of the motor, and select the highest motor health index as the health benchmark value.

5. The rotor life prediction method based on big data according to claim 1, characterized in that, Also includes: Determine the operating condition at the current moment. Based on the distance between the current operating condition parameters and the cluster centers of all clusters, select the operating condition corresponding to the nearest cluster as the operating condition at the current moment.

6. The rotor life prediction method based on big data according to claim 1, characterized in that, Clustering the aforementioned operating parameters yields various operating conditions, including: The K-Means algorithm is used to cluster the operating parameters to obtain multiple clusters corresponding to various operating conditions. The elbow method is used to determine the number of clusters.

7. The rotor life prediction method based on big data according to claim 1, characterized in that, Also includes: Before calculating the health degradation index of the motor at each time point, the time series of the motor health index is smoothed.

8. The rotor life prediction method based on big data according to claim 1, characterized in that, Narrowbands are constructed based on the characteristic frequencies of the peak values, including: A narrow band with a width of ±1Hz is constructed centered on each characteristic frequency.

9. The rotor life prediction method based on big data according to claim 4, characterized in that, The early operation phase is the period within a preset time frame after the motor's first operation.

10. A rotor life prediction system based on big data, comprising a processor and a memory, characterized in that, The memory stores a computer program, and the processor executes the computer program to implement the rotor life prediction method based on big data as described in any one of claims 1-9.

Images