CN117390974B

CN117390974B - Method, device and medium for predicting the remaining service life of a rotating device

Info

Publication number: CN117390974B
Application number: CN202311700644.7A
Authority: CN
Inventors: 窦浩; 戴林杉
Original assignee: KSB Shanghai Pump Co Ltd
Current assignee: KSB Shanghai Pump Co Ltd
Priority date: 2023-12-12
Filing date: 2023-12-12
Publication date: 2024-03-05
Anticipated expiration: 2043-12-12
Also published as: CN117390974A

Abstract

The present invention relates to a method, apparatus and medium for predicting the remaining useful life of a rotating device. The method comprises the following steps: acquiring acceleration detection data about each detection position of the rotating equipment and working condition data about the working condition of the rotating equipment; preprocessing the acquired acceleration detection data so as to generate time domain features and frequency domain features; generating a real-time vector corresponding to the current operating condition based on the generated time domain features and frequency domain features so as to calculate target health state data about the rotating equipment based on the real-time vector and the target reference vector; determining whether an inflection point appears in the historical health status data curve; and in response to the inflection point appearing, selecting a target machine learning model to fit a remaining useful life prediction curve to predict a remaining useful life of the rotating equipment. Therefore, the method and the device can accurately predict the residual service life of the rotating equipment in different running environments and decay periods.

Description

Method, device and medium for predicting the remaining service life of a rotating device

Technical Field

The present invention relates generally to industrial control, and in particular, to a method, computing device, and computer-readable storage medium for predicting a remaining useful life of a rotating device.

Background

The Remaining Useful Life (RUL) is the length of time that the device is running before repair or replacement is required. By estimating the remaining useful life of the equipment, the equipment owner or supplier can reasonably schedule the time to perform equipment repair or replacement in order to optimize equipment operating efficiency and avoid unplanned outages.

Conventional methods for predicting the remaining useful life of rotating equipment (e.g., without limitation, rotating machinery such as pumps, motors, etc.) are, for example: collecting the decay data of a large number of similar products of the rotating equipment as sample data, then establishing a mathematical model and a physical model by combining the sample data on the basis of failure mechanism analysis of the rotating equipment, combining the mathematical model and the physical model with data driving based on an information fusion method, extracting characteristics by adopting a data driving method on the basis of the mathematical model and the physical model of the rotating equipment, and then predicting the residual service life of the rotating equipment by combining methods such as a statistical theory, artificial intelligence and the like.

By adopting the scheme, after the inflection point of the health state trend curve is detected and determined, the target machine learning model is selected to be re-fitted with the residual service life prediction curve instead of adopting a single model, so that the target machine learning model determined by the method can be suitable for the residual service life prediction of the current rotating equipment and the whole decay period of the rotating equipment.

In the above conventional method, a large amount of degradation data of the same type of product of the rotating equipment needs to be collected for modeling. There is a significant difference due to the different rotating devices. Taking a pump as an example, the difference between the running environment, the working condition, the load and the like is large, and the stress and the pumping medium of the pump are possibly different, so that the fading trend of different pumps is not necessarily the same even if the pumps belong to the same type of products, therefore, a model constructed based on the fading data of the same type of products of the rotating equipment cannot be suitable for the current target rotating equipment, and a single model constructed based on the fading data of the same type of products of the rotating equipment cannot be suitable for the whole fading period of the rotating equipment. The inapplicability of the degradation data and the single model makes it difficult to accurately predict the remaining service life of the rotating equipment in different operating environments and degradation cycles according to the conventional method.

In summary, conventional methods for predicting the remaining useful life of rotating equipment cannot accurately predict the remaining useful life of rotating equipment in different operating environments and decay periods.

Disclosure of Invention

In view of the foregoing, the present invention provides a method, computing device, and computer-readable storage medium for predicting remaining useful life of a rotating device that enables accurate prediction of remaining useful life for rotating devices in different operating environments and decay periods.

According to a first aspect of the present invention there is provided a method for predicting the remaining useful life of a rotating equipment, the method comprising: acquiring acceleration detection data about each detection position of the rotating equipment and working condition data about the working condition of the rotating equipment; preprocessing the acquired acceleration detection data so as to generate time domain features and frequency domain features; generating a real-time vector corresponding to the current operating condition based on the generated time domain features and frequency domain features so as to calculate target health state data about the rotating equipment based on the real-time vector and the target reference vector; determining whether an inflection point appears in a historical health status data curve formed based on the cached target health status data; and in response to determining that the health status trend curve exhibits an inflection point, selecting a target machine learning model to fit a remaining useful life prediction curve so as to predict a remaining useful life of the rotating equipment.

In some embodiments, the method for predicting remaining useful life of a rotating equipment further comprises: caching target health status data about a rotating device generated within a predetermined period, the predetermined period being adjustable; and in response to determining that the health trend curve exhibits an inflection point, determining the cached target health data after the occurrence of the inflection point as health decay data for the rotating device.

In some embodiments, determining whether an inflection point appears in a historical state of health data curve formed based on the cached target state of health data comprises: generating a historical health status data curve based on the cached target health status data; dividing the historical health status data curve into a plurality of sliding windows; calculating an inclination angle of each sliding window in the plurality of sliding windows; and determining that an inflection point occurs at the current sliding window in response to determining that the tilt angle of the current sliding window exceeds the predetermined tilt angle threshold range.

In some embodiments, selecting the target machine learning model to fit the remaining useful life prediction curve to predict the remaining useful life of the rotating equipment comprises: responsive to determining that the health status trend curve exhibits inflection points, re-fitting the historical health status data curve with the plurality of candidate machine learning models to select a matched target machine learning model from the plurality of candidate machine learning models; training the determined target machine learning model based on the health decay data to update parameters of the target machine learning model; fitting the historical health state data curve by using the updated parameter target machine learning model so as to determine whether the fitting effect meets a preset condition; and in response to determining that the fitting effect meets a predetermined condition, predicting the remaining service life of the rotating equipment by using the fitted target machine learning model.

In some embodiments, predicting the remaining useful life of the rotating equipment using the fitted target machine learning model comprises: generating a residual service life prediction curve by adopting the fitted target machine learning model; determining an intersection point of the generated remaining life prediction curve and the specific health state data; and calculating the time length between the current time and the time corresponding to the intersection point so as to determine the residual service life of the rotating equipment based on the time length.

In some embodiments, the remaining life prediction curve is fitted using an exponential model whose parameters are calculated based on a least squares method.

In some embodiments, the rotating apparatus is a pump, the method further comprising: based on the current vibration intensity value data of the pump, initial health state data of the pump is determined.

In some embodiments, determining the initial health status data of the pump includes any one of: determining a vibration level of the pump based on the current vibration intensity value data of the pump, so as to determine initial health state data of the pump based on the vibration level of the pump; or determining initial health status data of the pump based on the current vibration intensity value data of the pump and a predetermined mapping algorithm.

In some embodiments, generating a real-time vector corresponding to the current operating condition based on the generated time domain features, frequency domain features includes: performing standardization and dimension reduction processing on time domain features and frequency domain features of working condition data of the current working condition so as to enable the variance contribution rate of the dimension reduced data to be larger than a preset variance threshold; and generating a real-time vector corresponding to the current working condition based on the dimension-reduced data.

In some embodiments, calculating the target health status data for the rotating device includes: confirming whether a reference vector of a working condition corresponding to the real-time vector exists in a reference vector library; in response to determining that no reference vector exists in the reference vector library for the operating condition corresponding to the real-time vector, determining a predicted reference vector generated based on a reference vector prediction model as a target reference vector matching the real-time vector; in response to determining that a reference vector of a working condition corresponding to a real-time vector exists in a reference vector library, determining the reference vector of the working condition corresponding to the real-time vector as a target reference vector matched with the real-time vector; calculating the distance between the real-time vector and the target reference vector; and generating target health status data about the rotating device based on the calculated distance; based on the determined initial health state data, a correction is made for the calculated target health state data about the rotating equipment to obtain corrected target health state data.

In some embodiments, the reference vector predictive model is generated via: clustering is carried out on the real-time vectors under each working condition in different working conditions so as to determine a reference vector corresponding to each working condition based on a clustering center; and generating a reference vector prediction model by using a multiple linear regression model based on the working condition data cached by the reference vector library and the reference vector corresponding to the working condition.

In some embodiments, the acceleration detection data comprises: acceleration detection data of the pump driving end bearing in the vertical direction, acceleration detection data of the pump driving end bearing in the horizontal direction and acceleration detection data of the pump driving end bearing in the axial direction, wherein the working condition data is at least one of temperature, rotating speed and flow; the time domain features include: regarding a plurality of effective values, peaks, kurtosis coefficients of the acceleration detection data, the frequency domain features include at least one of: the inner ring fault frequency energy, the outer ring fault frequency energy, and the pump drive end and non-drive end vertical direction phase differences of the pump are related. According to a second aspect of the present invention there is provided a computing device comprising: at least one processing unit; at least one memory coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, the instructions when executed by the at least one processing unit cause the apparatus to perform the steps of the method according to the first aspect.

According to a third aspect of the present invention, there is provided a computer readable storage medium having stored thereon a computer program which, when executed by a machine, implements a method according to the first aspect.

The summary is provided to introduce a selection of concepts in a simplified form that are further described below in the detailed description. This summary is not intended to identify key features or essential features of the invention, nor is it intended to be used to limit the scope of the invention.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present invention, and that other drawings can be obtained according to the provided drawings without inventive effort for a person skilled in the art.

Fig. 1 shows a schematic diagram of a system implementing a method for predicting the remaining useful life of a rotating equipment according to an embodiment of the present invention.

Fig. 2 shows a flow chart of a method for predicting the remaining service life of a rotating device according to an embodiment of the invention.

Fig. 3 illustrates a flowchart of a method for calculating target health status data for a rotating device, according to an embodiment of the invention.

FIG. 4 illustrates a flowchart of a method for generating a real-time vector corresponding to a current operating condition according to an embodiment of the present invention.

Fig. 5 illustrates a flow chart of some methods for determining remaining useful life with respect to a rotating equipment, according to an embodiment of the present invention.

FIG. 6 illustrates a flowchart of a method for generating a real-time vector corresponding to a current operating condition according to an embodiment of the present invention.

Fig. 7 illustrates a flowchart of a method of generating a reference vector predictive model in accordance with an embodiment of the invention.

Fig. 8 schematically illustrates a block diagram of a system implementing a method for predicting the remaining useful life of a rotating equipment according to further embodiments of the present invention.

Fig. 9 illustrates an effect diagram of water pump health indication data determined by a method for predicting remaining service life of a rotating equipment according to an embodiment of the present invention.

Fig. 10 schematically shows a block diagram of an electronic device suitable for use in implementing embodiments of the invention.

Detailed Description

Preferred embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While the preferred embodiments of the present disclosure are illustrated in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

The term "comprising" and variations thereof as used herein means open ended, i.e., "including but not limited to. The term "or" means "and/or" unless specifically stated otherwise. The term "based on" means "based at least in part on". The terms "one example embodiment" and "one embodiment" mean "at least one example embodiment. The term "another embodiment" means "at least one additional embodiment". The terms "first," "second," and the like, may refer to different or the same object.

As described above, the conventional method for predicting the remaining service life of the rotating equipment cannot accurately predict the remaining service life of the rotating equipment in different operating environments and degradation cycles.

To at least partially address one or more of the above-mentioned problems, as well as other potential problems, the present invention proposes a method for predicting the remaining useful life of a rotating device. In the scheme of the invention, acceleration detection data of each detection position of the rotating equipment and working condition data of the working condition of the rotating equipment are obtained; then, preprocessing the acquired acceleration detection data so as to generate time domain features and frequency domain features; and generating a real-time vector corresponding to the current working condition based on the generated time domain feature and the frequency domain feature so as to calculate the target health state data about the rotating equipment based on the real-time vector and the target reference vector. Additionally, determining whether an inflection point appears in a historical health status data curve formed based on the cached target health status data; and in response to determining that the health state trend curve has an inflection point, selecting a target machine learning model to fit a residual service life prediction curve so as to predict the residual service life of the rotating equipment. In addition, when the inflection point of the health state trend curve is determined, the target machine learning model is selected to fit the residual service life prediction curve, and the characteristics that the health state trend curve of different regression stages shows different changes are taken into consideration to select the proper target machine learning model for re-fitting the residual service life prediction curve, so that the defect that a single model is not suitable for the whole regression period of the rotating equipment is overcome. Therefore, the method and the device can accurately predict the residual service life of the rotating equipment in different running environments and decay periods.

The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only some embodiments of the present invention, but not all embodiments, and they should not be construed as limiting the protection scope of the present application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

Fig. 1 shows a schematic diagram of a system 100 implementing a method for predicting the remaining useful life of a rotating equipment according to an embodiment of the present invention. As shown in fig. 1, the system 100 includes: a computing device 110, a rotating device 130, an acceleration detection means 140. In some embodiments, the computing device 110, the rotation device 130, and the acceleration detection apparatus 140 interact with data directly or via a network.

With respect to the rotary device 130, in some embodiments, it is for example and without limitation a variable frequency drive, or a power frequency driven rotary device, such as a variable frequency driven pump. In other embodiments, it is for example, but not limited to, an electric machine.

With respect to the computing device 110, it is used to predict the remaining useful life of the rotating device. Specifically, the computing device 110 is configured to obtain acceleration detection data about each detection position of the rotating device and operating condition data about an operating condition of the rotating device; and preprocessing the acquired acceleration detection data so as to generate time domain features and frequency domain features. The computing device 110 is further configured to generate a real-time vector corresponding to the current operating condition based on the generated time-domain features, frequency-domain features, so as to calculate target health status data about the rotating device based on the real-time vector and the target reference vector; determining whether an inflection point appears in a historical health status data curve formed based on the cached target health status data; and in response to determining that the health status trend curve exhibits an inflection point, selecting a target machine learning model to fit a remaining useful life prediction curve so as to predict a remaining useful life of the rotating equipment. In some embodiments, computing device 110 may have one or more processing units, including special purpose processing units such as GPUs, FPGAs, and ASICs, as well as general purpose processing units such as CPUs. In addition, one or more virtual machines may also be running on each computing device. The computing device 110 includes, for example: acceleration detection data and operating condition data acquisition unit 112, time domain feature and frequency domain feature generation unit 114, target health state data generation unit 116, inflection point determination unit 118, remaining life prediction unit 120. The acceleration detection data and operating condition data acquisition unit 112, the time domain feature and frequency domain feature generation unit 114, the target health state data generation unit 116, the inflection point determination unit 118, and the remaining life prediction unit 120 may be configured on one or more computing devices 110.

And an acceleration detection data and condition data acquisition unit 112 for acquiring acceleration detection data on each detection position of the rotating apparatus and condition data on the condition in which the rotating apparatus is operated.

Regarding the time domain feature and frequency domain feature generating unit 114, it is used for preprocessing the acquired acceleration detection data so as to generate a time domain feature and a frequency domain feature.

The target health state data generation unit 116 is configured to generate a real-time vector corresponding to the current operating condition based on the generated time domain feature and frequency domain feature, so as to calculate target health state data about the rotating device based on the real-time vector and the target reference vector.

Regarding the inflection point determining unit 118, it is used to determine whether an inflection point appears in a historical health status data curve formed based on the buffered target health status data.

Regarding the remaining service life prediction unit 120, it is used for responding to the determined health status trend curve and appearing the inflection point, choose the goal machine learning model to fit the remaining service life prediction curve, in order to predict the remaining service life of the rotating equipment.

Fig. 2 shows a flow chart of a method 200 for predicting the remaining useful life of a rotating equipment according to an embodiment of the present invention. It should be appreciated that the method 200 may be performed at the electronic device 1000 depicted in fig. 10, for example. May also be performed at the computing device 110 depicted in fig. 1. It should be appreciated that method 200 may also include additional actions not shown and/or may omit actions shown, the scope of the invention being not limited in this respect.

At step 202, the computing device 110 obtains acceleration detection data for each detected position of the rotating device and operating condition data for an operating condition of the rotating device.

As for the rotating device, it is, for example, but not limited to, a variable frequency driven pump, motor, or a power frequency driven pump, motor, or the like.

Acceleration detection data concerning each detection position of the rotating device is, for example, acceleration waveform data acquired via an acceleration detection means (for example, an acceleration sensor) for each detection position of the rotating device. In an embodiment in which the rotating device is a pump, the acceleration detection data for each detection position includes, for example: acceleration detection data of the vertical direction of the pump driving end bearing, acceleration detection data of the horizontal direction of the pump driving end bearing and acceleration detection data of the axial direction data of the pump driving end bearing.

The operating state data about the operating state of the rotating device, for example, are temperature, rotational speed, flow rate, etc. In some embodiments, the operating condition data for the operating condition of the rotating equipment may be one or more of the data described above.

At step 204, the computing device 110 pre-processes the acquired acceleration detection data to generate time domain features, frequency domain features. For example, the computing device 110 performs data quality monitoring and feature extraction on the acquired acceleration detection data to ultimately generate time domain features, frequency domain features for each detection location.

Regarding a method of generating a time domain feature, a frequency domain feature, it includes, for example: the computing device 110 determines whether the acquired acceleration detection data has an anomaly, the anomaly including at least one of: the loss rate of the acceleration detection data exceeds a preset loss rate threshold value, the difference value of the acceleration detection data is abnormal, and the variance of the acceleration detection data is abnormal; and extracting features for the acceleration detection data if it is determined that the acquired acceleration detection data is not abnormal, so as to generate time domain features and frequency domain features. And if the obtained current acceleration detection data is determined to be abnormal, deleting the current acceleration detection data with the abnormality.

Regarding time domain features, it includes, for example: regarding a plurality of items of the effective value, peak value, kurtosis coefficient of the acceleration detection data (waveform data and pulse data). In some embodiments, the time domain features further comprise: and a plurality of peak-to-peak value, variance and standard deviation of the acceleration detection data.

The effective value (RSM), which is also called a root mean square value, is a square root of the acceleration detection data at a time mean square value, which is also called a mean square amplitude value, reflects the overall vibration level of the signal, is relatively sensitive to the gear rotation imbalance fault, but cannot reflect the signal number isolated peak, and therefore, the effective value (RSM) cannot detect the early minor fault. The algorithm for generating the effective value (RSM) in the time domain feature is described below in conjunction with equation (1):

（1）

In the above-mentioned formula (1),representing the significant values included in the time domain features. N represents the number of acceleration detection data. />Representing the nth acceleration detection data.

Regarding Kurtosis (Kurtosis) coefficient, which is a numerical statistic reflecting the distribution characteristics of vibration signals (amplitudes), it can be used to describe the probability that a fault causes a large pulse. The kurtosis coefficient is independent of bearing rotation speed, size, load and the like of the rotating equipment, but is particularly sensitive to impact signals, so that the method is particularly suitable for surface damage faults, and particularly is sensitive to early faults of the bearing. If the bearings of the rotating equipment fail, the kurtosis coefficient value increases.

Regarding the peak value, it is the maximum value of the absolute value of the amplitude of the acceleration detection data over one time slice. As with the peak-to-peak value, the peak value describes the range of the acceleration detection data amplitude.

Regarding Peak-to-Peak (Peak), which is the difference of Peak-to-Peak values between the maximum positive value and the minimum negative value of the amplitude of the acceleration detection data over a period of time, the Peak-to-Peak value describes the range of variation of the signal value. The algorithm for generating the peak-to-peak value in the time domain feature is described below in conjunction with equation (2):

（2）

in the above-mentioned formula (2), Representing the peak-to-peak value included in the time domain feature. />Representing the nth acceleration detection data. />Representing the maximum positive value of the amplitude of the acceleration detection data over a period of time. />Representing the smallest negative value of the amplitude of the acceleration detection data over a period of time.

With respect to variance, it is the average of the squares of the amplitude and mean deviation over a period of time of the acceleration detection data. The variance reflects the magnitude of fluctuation of the acceleration detection data and represents the intensity of the alternating current component in the acceleration detection data. The algorithm for generating the variance in the time domain features is described below in conjunction with equation (3):

（3）

regarding the standard deviation, which is the square root of the variance, the stability of the acceleration detection data is described, and compared with the variance, the standard deviation can more intuitively reflect the intensity of fluctuation of the acceleration detection data. The algorithm for generating standard deviations in the time domain features is described below in conjunction with equation (4):

（4）

in the above formulas (3) and (4),representing the variance included in the time domain features. />Representing the standard deviation included in the time domain features. />Representing the nth acceleration detection data. N represents the number of acceleration detection data.

Regarding the frequency domain features, it includes, for example, at least one of the following: the inner ring fault frequency energy, the outer ring fault frequency energy, and the pump drive end and non-drive end vertical direction phase differences of the pump are related.

At step 206, the computing device 110 generates a real-time vector corresponding to the current operating condition based on the generated time-domain features, frequency-domain features, to calculate target health data for the rotating device based on the real-time vector and the target reference vector.

Regarding a method of generating a real-time vector corresponding to a current operating condition, it includes, for example: the computing equipment 110 performs standardization and dimension reduction processing on time domain features and frequency domain features of working condition data of the current working condition so as to enable the variance contribution rate of the dimension reduced data to be larger than a preset variance threshold; and generating a real-time vector corresponding to the current working condition based on the dimension-reduced data. The method 600 for generating the real-time vector corresponding to the current operating condition will be described in detail below with reference to fig. 6, and will not be described herein.

As regards a method of calculating target health status data about a rotating device, it comprises, for example: the computing device 110 confirms whether there is a reference vector of the operating condition corresponding to the real-time vector in the reference vector library; in response to determining that no reference vector exists in the reference vector library for the operating condition corresponding to the real-time vector, determining a predicted reference vector generated based on a reference vector prediction model as a target reference vector matching the real-time vector; in response to determining that a reference vector of a working condition corresponding to a real-time vector exists in a reference vector library, determining the reference vector of the working condition corresponding to the real-time vector as a target reference vector matched with the real-time vector; calculating the distance between the real-time vector and the target reference vector; and generating target health status data about the rotating device based on the calculated distance; based on the determined initial health state data, a correction is made for the calculated target health state data about the rotating equipment to obtain corrected target health state data. The method 300 for generating the target health status data about the rotating device will be described in detail below with reference to fig. 3, and will not be described here again.

At step 208, the computing device 110 determines whether an inflection point appears in a historical state of health data curve formed based on the cached target state of health data. If the computing device 110 determines that the health status trend curve does not exhibit an inflection point, it waits at step 208 to continue to determine if the historical health status data curve formed based on the target health status data exhibits an inflection point.

It should be appreciated that the computing device 110 determines whether an inflection point indicating health degradation occurs based on the target health state data calculated by the real-time vector under the current working condition of the current rotating device and the historical health state data curve generated based on the cached target health state data of the current rotating device itself, rather than degradation data of other rotating devices of the same kind of products (the operating environments and working conditions of the other rotating devices of the same kind of products may be different from those of the rotating device to be currently predicted), so the present invention can make up for the disadvantage that the model built based on the same kind of products cannot be applied to the current specific rotating device, and can effectively and accurately determine the inflection point of health degradation for the rotating devices of different operating environments.

Regarding a method of determining whether an inflection point appears in a historical state of health data curve, it includes, for example: computing device 110 generates a historical state of health data curve based on the cached target state of health data; dividing the historical health status data curve into a plurality of sliding windows; calculating an inclination angle of each sliding window in the plurality of sliding windows; determining whether the inclination angle of the current sliding window exceeds a preset inclination angle threshold range; and determining that an inflection point occurs at the current sliding window in response to determining that the tilt angle of the current sliding window exceeds the predetermined tilt angle threshold range. The method 400 for determining whether the inflection point appears in the historical health status data curve will be described in detail with reference to fig. 4, and will not be described here.

At step 210, if the computing device 110 determines that the health trend curve exhibits an inflection point, a target machine learning model is selected to fit the remaining life prediction curve in order to predict the remaining life of the rotating device. For example, FIG. 9 shows a schematic effect of a historical health data curve according to an embodiment of the invention. In fig. 9, the abscissa is the time corresponding to the number of samples sampled, for example, the abscissa is 2000, which indicates the time corresponding to the 2000 th sample. The numerical unit of the abscissa is, for example, "sampling interval time". The specific value of time corresponding to the abscissa 2000 is the product of 2000 and the sampling interval time. The unit of sampling interval time is hours. The ordinate is historical health data formed based on the cached target health data. The unit is, for example, "score". The marker 910 indicates a historical health data curve determined by the computing device 110. The label 914 indicates the remaining life prediction curve re-fitted by selecting the target machine learning model from the plurality of candidate machine learning models. The flag 930 indicates particular health status data (in some embodiments, the particular health status data is, for example, 0, in other embodiments, the particular health status data is, for example, 20 or other values). Computing device 110 determines an intersection of generated remaining life prediction curve 914 with specific health state data 930 (e.g., the time indicated by the intersection is X2); and calculating a time length (e.g., a time length X2-X1) between the current time (e.g., a time corresponding to the current time is X1) and the time corresponding to the intersection point, so as to determine a remaining service life of the rotating apparatus based on the time length.

By adopting the means, after the inflection point of the health state trend curve is detected and determined, the target machine learning model is selected to be re-fitted with the residual service life prediction curve instead of adopting a single model, so that the target machine learning model determined by the method can be suitable for the residual service life prediction of the current rotating equipment and the whole decay period of the rotating equipment.

It has been found that the decay function of a rotating device, such as a pump, is a piecewise function. That is, the functional form of the health trend curve may change at different stages. For example, the front phase of the state of health trend curve is natural wear, the state of health trend curve of the phase corresponds to a linear function, for example, and the rear phase enters a decay phase, for example, the state of health trend curve of the phase corresponds to another function (for example, but not limited to, an exponential function, for example). Thus, there are different stages of target machine learning models for fitting the remaining life prediction curve. Thus, determining the inflection point of the health trend curve requires selecting an appropriate target machine learning model to re-fit the remaining life prediction curve.

A method for predicting the remaining service life of a rotating device, for example, comprising: if the computing device 110 determines that the health trend curve exhibits inflection points, re-fitting the historical health data curve with the plurality of candidate machine learning models to select a matched target machine learning model from the plurality of candidate machine learning models; training the determined target machine learning model based on the health decay data to update parameters of the target machine learning model; fitting the historical health state data curve by using the updated parameter target machine learning model so as to determine whether the fitting effect meets a preset condition; and in response to determining that the fitting effect meets a predetermined condition, predicting the remaining service life of the rotating equipment by using the fitted target machine learning model.

For example, first, the computing device 110 re-fits the historical state of health data curve with the plurality of candidate machine learning models in response to determining that the state of health trend curve exhibits inflection points, to select a matching target machine learning model among the plurality of candidate machine learning models.

By adopting the means, after the inflection point of the health state trend curve is detected and determined, the model fitting effect is used for adjusting the type of the target machine learning model instead of a single model, so that the target machine learning model selected by the method can be suitable for the current rotating equipment and the whole decay period of the rotating equipment.

For example, the computing device 110 determines the fitting effect of the different candidate machine learning models based on a mean-square error (MSE) to select the target machine learning model with the best fitting effect. The method of calculating the MSE is described below in conjunction with equation (5):

（5）

in the above-mentioned formula (5),representing the actual value of the health status data corresponding to the ith sample on the historical health status data curve. />And estimating the health state data corresponding to the ith sample predicted by the candidate machine learning model.Representing the mean square error between the candidate machine learning model f and the historical state of health data curve D. m represents the number of samples。

Second, the computing device 110 trains the determined target machine learning model based on the health decay data to update parameters of the target machine learning model.

It should be appreciated that by training the selected target machine learning model using the health decay data of the current rotational device (e.g., pump) itself, the present invention can determine in real time whether to update the parameters of the selected target machine learning model based on the fitting effect of the health decay data itself, thereby making the parameters of the target machine learning model more suitable for the prediction of the remaining useful life of the decay cycle in which the current rotational device is located. Moreover, as the amount of decay data behind the buffered inflection point increases, the better the fit of the target machine learning model trained based on the decay data will be.

Furthermore, the computing device 110 fits the historical health data curve with the updated parameter target machine learning model to determine whether the fit meets a predetermined condition. If the computing device 110 determines that the fitting effect does not meet the predetermined condition, continuing to update the parameters of the target machine learning model to re-fit the remaining life prediction curve using the target machine learning model after the parameters are adjusted until the fitting effect is determined to meet the predetermined condition.

Regarding the method of calculating the fitting effect, it is based on, for example, a mean-square error (MSE) algorithm. For example, if the computing device 110 determines that a fitting effect calculated based on a mean-square error (MSE) is less than a predetermined threshold, then it is determined that the fitting effect satisfies a predetermined condition.

Regarding the remaining life prediction curve, it is fitted, for example, using an exponential model whose parameters are calculated, for example, based on the least square method. The following describes the calculation method of the exponential model in conjunction with equation (6):

（6）

in the above-mentioned formula (6),representing health status data corresponding to the ith sample on the curve.Three parameters representing the exponential model, respectively, are obtained according to least squares learning. / >Representing the ith sample on the curve.

Finally, if the computing device 110 determines that the fit effect meets the predetermined condition, the fitted target machine learning model is used to predict the remaining useful life of the rotating device. For example, computing device 110 generates a remaining useful life prediction curve using the fitted target machine learning model; determining an intersection point of the generated remaining life prediction curve and the specific health state data; and calculating the time length between the current time and the time corresponding to the intersection point so as to determine the residual service life of the rotating equipment based on the time length.

By adopting the means, the invention can calculate the accurate target health degree based on the real-time vector of the current rotating equipment and the target reference vector. In addition, the invention utilizes a historical health state data curve generated based on the cached actual target health state data of the current rotating equipment to determine whether an inflection point indicating health decay occurs or not, instead of utilizing decay data of similar products; furthermore, when the inflection point of the health state trend curve is determined, the target machine learning model is selected to fit the residual service life prediction curve, and the characteristics that the health state trend curve of different regression stages shows different changes can be considered to select the proper target machine learning model for re-fitting the residual service life prediction curve, so that the defect that a single model is not suitable for the whole regression period of the rotating equipment is overcome. Therefore, the method and the device can accurately predict the residual service life of the rotating equipment in different running environments and decay periods.

Fig. 3 illustrates a flow chart of a method 300 for calculating target health status data for a rotating device, in accordance with an embodiment of the present invention. It should be appreciated that the method 300 may be performed, for example, at the electronic device 1000 depicted in fig. 10. May also be performed at the computing device 110 depicted in fig. 1. It should be appreciated that method 300 may also include additional actions not shown and/or may omit actions shown, the scope of the invention being not limited in this respect.

At step 302, computing device 110 confirms whether there is a reference vector for the operating condition corresponding to the real-time vector in the reference vector library.

For example, the reference vector library stores reference vectors corresponding to a plurality of different working conditions (such as different temperatures, rotational speeds and flow rates). Computing device 110 may query a reference vector library to see if there is a reference vector for the operating conditions corresponding to the current real-time vector.

At step 304, if computing device 110 determines that there is no reference vector in the reference vector library for the operating conditions corresponding to the real-time vector, a predicted reference vector generated based on the reference vector prediction model is determined to be a target reference vector that matches the real-time vector.

At step 306, if computing device 110 determines that there is a reference vector for the operating condition corresponding to the real-time vector in the reference vector library, the reference vector for the operating condition corresponding to the real-time vector is determined to be a target reference vector that matches the real-time vector.

At step 308, the computing device 110 calculates a distance between the real-time vector and the target reference vector.

With respect to a method for calculating a distance between a real-time vector and a target reference vector, it includes, for example: the computing device 110 calculates a distance between the real-time vector and the target reference vector based on a euclidean distance algorithm. It should be appreciated that other algorithms may be employed by the present invention to calculate the distance between the real-time vector and the target reference vector in real-time. The algorithm for calculating the distance between the real-time vector and the target reference vector is described below in conjunction with equation (7):

（7）

in the above-mentioned formula (7),representing the euclidean distance between the real-time vector and the target reference vector. />Representing the ith data in the real-time vector. />Representing the ith data in the target reference vector. n represents the number of data included in the real-time vector and the target reference vector.

At step 310, the computing device 110 generates target health status data for the rotating device based on the calculated distance.

It will be appreciated that the greater the distance between the real-time vector and the target reference vector, the greater the likelihood that the real-time vector indicating the current operating condition of the rotating equipment will deviate from the normal target reference vector. Based on the generated target health assessment data, the user is presented with a visual understanding of the health of the rotating equipment (e.g., without limitation, a power frequency driven pump).

Regarding a method of generating target health status data, it for example comprises: the computing device 110 converts the calculated distance into health indication data regarding the health status of the rotating device based on the adjustment function; and denoising the health degree indicating data by using a denoising function to acquire target health state evaluation data about the rotating equipment.

With respect to the adjustment function, it is used to cause the input distance and the output health indication data to exhibit an opposite direction change. The input data of the adjustment function are: a distance between the calculated real-time vector and the target reference vector, a predetermined distance threshold; the output data of the adjustment function is: health indication data regarding the health status of the rotating device. In some embodiments, the health indication data for the initial state is, for example and without limitation, 100. The following describes the functional expression of the adjustment function in conjunction with equation (8):

（8）

in the above-mentioned formula (8),representing the calculated distance, which is used to input the adjustment function. />Representing a predetermined distance threshold. It should be appreciated that the predetermined distance threshold is obtained via pre-training. />Representing health indication data about the health status of the rotating device output via the adjustment function.

Regarding the noise reduction function, it is used for noise reduction for the health indication data. Since, in a real situation, the acceleration detection device may have an unstable problem, and sometimes the detection data of the acceleration detection device may have an error, and the health condition of the apparatus may not be correctly reflected. For example, the acceleration detection device may collect the mutation of the detection data caused by the user clicking the equipment to patrol the rotating equipment, but the mutation of the detection data may not be filtered through the data quality monitoring, and the mutation of the detection data may not truly reflect the health condition of the equipment, if the health degree indication data is calculated based on the data, the result may be inaccurate, so that the problem of inaccurate detection result caused by the above reasons can be avoided by using the noise reduction function to reduce the noise of the health degree indication data. In some embodiments, the noise reduction function may be constructed based on a moving average algorithm. The manner of calculation of the noise reduction function is illustrated below in conjunction with equation (9):

（9）

in the above-mentioned formula (9),representing target health status assessment data regarding the rotating equipment. />Representing the nth health indication data. / >Represents->Personal health indication data. T represents a predetermined number.

At step 312, the computing device 110 corrects the calculated target health state data for the rotating device based on the determined initial health state data to obtain corrected target health state data.

In some embodiments, the rotating device is a pump. As to a method of correcting for target health state data, it includes, for example: the computing device 110 determines initial health status data for the pump based on the current vibration intensity value data for the pump; and based on the determined initial health status data, correcting the calculated target health status data about the rotating equipment. The vibration intensity value data of the pump is an effective value of the vibration speed of the pump, that is, RMS, which is a characteristic quantity describing the vibration state of one pump.

It should be appreciated that the current pump may not be a best-health pump, the base state of the pumps may be different, and the value indicated by the starting health state data for the target health state data for the rotating equipment may also be different (e.g., if the best health state of the pump is indicated at 100 points, if the current pump to be predicted is not a best-health pump, its starting health state data may not be 100 points), and therefore it may be necessary to determine the initial health state data for the current pump to be predicted.

Regarding the method of determining initial health status data for a pump, it may include a variety of ways, for example, the computing device 110 determining the class to which the pump belongs based on its center height and rotational speed; and determining initial health state data based on the vibration intensity value data and the belonging category.

In some embodiments, the computing device 110 determines a vibration level of the pump based on the current vibration intensity value data of the pump, so as to determine initial health state data of the pump based on the vibration level of the pump. Specifically, for example, if the computing device 110 determines that the RMS value of the pump is less than 4.5mm/s, then the pump is determined to be in a low risk state, the corresponding vibration level of the pump is, for example, "a", the computing device 110 causes the initial health state data of the pump to be less than or equal to 100 minutes and greater than or equal to 80 minutes, wherein the greater the RMS value of the pump, the lower the initial health state data of the determined pump. If the computing device 110 determines that the RMS value of the pump is equal to 4.5mm/s, the initial health data for the pump is 80 minutes. If the computing device 110 determines that the RMS value of the pump is greater than 4.5mm/s and less than 11.2mm/s, then the pump is considered to be in a risky state, the corresponding vibration level of the pump is, for example, "B", and the computing device 110 causes the initial health state data of the pump to be less than 80 minutes and greater than or equal to 60 minutes, wherein the greater the RMS value of the pump, the lower the initial health state data of the determined pump. If the computing device 110 determines that the RMS value of the pump is equal to 11.2mm/s, the initial health status data for the pump is 60 minutes. If the computing device 110 determines that the RMS value of the pump is greater than 11.2mm/s, then the pump is considered to be in a high risk state, the vibration level of the corresponding pump is, for example, "C", and the computing device 110 causes the initial health state data of the pump to be less than 60 minutes and greater than or equal to 0 minutes, wherein the greater the RMS value of the pump, the lower the initial health state data of the determined pump. If the computing device 110 determines that the RMS value of the pump is greater than or equal to 45mm/s, the corresponding vibration level of the pump is, for example, "D", the computing device 110 causes the initial health status data of the pump to be 0 minutes.

In other embodiments, the computing device 110 determines the initial health data for the pump based on the current vibration intensity value data for the pump and a predetermined mapping algorithm. The calculation method of the predetermined mapping algorithm is described below in conjunction with equation (10):

（10）

in the above formula (10), x represents the current vibration intensity value data of the pump.Representing the determined initial health status data of the pump.

Regarding a method of correcting for target health state data based on initial health state data, it includes, for example: the computing device 110 subtracts the determined initial health state data from the calculated target health state data about the rotating device to obtain revised target health state data.

By adopting the means, the method can determine the reference vector matched with the corresponding working condition of the real-time vector of the current equipment, thereby being capable of accurately determining the target health data of the equipment under different working conditions; moreover, by correcting the calculated target health state data based on the determined initial health state data, the invention can accurately predict the residual service life of equipment in different initial states.

FIG. 4 illustrates a flowchart of a method 400 for generating a real-time vector corresponding to a current operating condition in accordance with an embodiment of the present invention. It should be appreciated that the method 400 may be performed, for example, at the electronic device 1000 depicted in fig. 10. May also be performed at the computing device 110 depicted in fig. 1. It should be appreciated that method 400 may also include additional actions not shown and/or may omit actions shown, the scope of the invention being not limited in this respect.

At step 402, the computing device 110 generates a historical state of health data curve based on the cached target state of health data. As shown, the computing device 110 generates a historical state of health data curve 910 based on the cached target state of health data.

At step 404, the computing device 110 divides the historical health data curve into a plurality of sliding windows.

At step 406, computing device 110 calculates a tilt angle for each of the plurality of sliding windows. For example, the computing device 110 calculates the inclination of the health status data at each sliding window termination time to the horizontal line of the health status data link at the sliding window termination time.

At step 408, computing device 110 determines whether the tilt angle of the current sliding window exceeds a predetermined tilt angle threshold range. If computing device 110 determines that the tilt angle of the current sliding window does not exceed the predetermined tilt angle threshold range, at step 412, it is determined that no inflection point occurs at the current sliding window. For example, computing device 110 determines that the inclination of the health data at the end of sliding window 920 to the health data link at the end of the sliding window (as indicated by reference 922) relative to the horizontal link is less than a predetermined inclination threshold range, then determines that no inflection point is present at sliding window 922.

At step 410, if computing device 110 determines that the tilt angle of the current sliding window is outside of the predetermined tilt angle threshold range, it is determined that an inflection point occurs at the current sliding window.

With respect to the predetermined tilt threshold range of the current sliding window, it is for example the upper and lower limits [ u-3σ, u+3σ ] of the tilt of the current sliding window](wherein,representing the standard deviation of the tilt angle of the current sliding window. u represents the average value of the tilt angle of the current sliding window).

If the computing device 110 determines that the tilt angle of the current sliding window exceeds the predetermined tilt angle threshold range, the slope of the historical health data curve at the current sliding window is excessively changed, i.e., it is determined that an inflection point occurs at the current sliding window. For example, the computing device 110 determines that an inflection point occurred at the sliding window 924 if the inclination of the health data link (as indicated by reference 926) at the end of the sliding window 924 with the health data link at the end of the sliding window relative to the horizontal is greater than a predetermined inclination threshold range. For example, inflection point 912.

By adopting the means, the inflection point of the change of the declining phase of the rotating equipment can be accurately determined.

Fig. 5 illustrates a flowchart of a method 500 for determining remaining useful life with respect to a rotating equipment, in accordance with an embodiment of the present invention. It should be appreciated that the method 500 may be performed, for example, at the electronic device 1000 depicted in fig. 10. May also be performed at the computing device 110 depicted in fig. 1. It should be appreciated that method 500 may also include additional actions not shown and/or may omit actions shown, the scope of the invention being not limited in this respect.

At step 502, if the computing device 110 determines that the health trend curve exhibits inflection points, the historical health data curve is re-fitted with the plurality of candidate machine learning models to select a matching target machine learning model from the plurality of candidate machine learning models. As shown in fig. 9, for example, computing device 110 determines that health trend curve 910 exhibits inflection point 912. Then, computing device 110 re-fits the historical health data curve using the plurality of candidate machine learning models; and determining fitting effects of the different candidate machine learning models based on a mean-square error (MSE) so as to select a target machine learning model with the best fitting effect.

At step 504, the computing device 110 trains the determined target machine learning model based on the health decay data to update parameters of the target machine learning model.

At step 506, the computing device 110 fits the historical health data curve with the target machine learning model of updated parameters to determine whether the fit effect meets a predetermined condition.

At step 508, if the computing device 110 determines that the fit meets the predetermined condition, a remaining life prediction curve is generated using the fitted target machine learning model. As shown in fig. 9, for example, computing device 110 generates a remaining useful life prediction curve 914 using the fitted target machine learning model.

At step 510, computing device 110 determines an intersection of the generated remaining life prediction curve with the particular health state data. For example, the computing device 110 determines the intersection of the generated remaining life prediction curve 914 with the particular health state data 930, such as the time indicated by X2 in fig. 9.

At step 512, the computing device 110 calculates a time period between the current time and the time corresponding to the intersection point to determine a remaining life time for the rotating device based on the time period. For example, the computing device 110 calculates a duration between the current time X2 and the time X1 corresponding to the intersection point.

The algorithm for calculating the remaining useful life is described below in conjunction with equation (11):

（11）

in the above-mentioned formula (11),representing the calculated remaining useful life. />Representing the time corresponding to the intersection point of the generated residual service life prediction curve and the specific health degree. />Representing the current time.

With respect to a particular degree of health, it is for example, but not limited to, "0".

By adopting the means, the method and the device can reselect the type of the target machine learning model according to the model fitting effect after detecting the change inflection point of the historical health state data curve; in addition, for the selected matched model, whether to update the model parameters is judged by utilizing real-time incremental fading data according to the fitting effect of the data, so that the model parameters and the model types can be more effectively applied to the prediction of the residual service life of the whole fading period of the rotating equipment.

FIG. 6 illustrates a flowchart of a method 600 for generating a real-time vector corresponding to a current operating condition in accordance with an embodiment of the present invention. It should be appreciated that the method 600 may be performed, for example, at the electronic device 1000 depicted in fig. 10. May also be performed at the computing device 110 depicted in fig. 1. It should be appreciated that method 600 may also include additional actions not shown and/or may omit actions shown, the scope of the invention being not limited in this respect.

At step 602, the computing device 110 performs normalization and dimension reduction processing on the time domain features and the frequency domain features of the operating mode data of the current operating mode so that the variance contribution rate of the dimension reduced data is greater than a predetermined variance threshold.

Regarding the method of performing the normalization processing, the normalization of data may be performed based on the mean (mean) and standard deviation (standard deviation) of the weighted features. For example, and without limitation, standard deviation normalization (Z-score normalization) methods are employed to map the weighted features to a distribution with a mean of 0 and a standard deviation of 1. The following illustrates the standardized manner in connection with equation (12):

（12）

in the above-mentioned formula (12),representing the normalized data. / >Representing weighted features prior to normalization. />Representing the mean of the weighted features prior to normalization. />Representing the standard deviation of the weighted features prior to normalization.

As regards the predetermined variance threshold, it is for example, but not limited to, 80%. It should be appreciated that too high a dimension is prone to interference with subsequent distance calculations. In addition, the dimension-reduced data is screened by the size of the variance contribution ratio, so that the interference of the dimension on the subsequent distance calculation can be reduced while the main characteristic component of the characteristic information is maintained.

The manner of the dimension reduction processing includes, for example: the computing device 110 performs dimension reduction on the high-dimensional data of the data subjected to the normalization processing based on the PCA (Principal Component Analysis) dimension reduction algorithm so as to extract a principal feature component in the high-dimensional data, and the variance contribution ratio of the extracted principal feature component is greater than a predetermined variance threshold. It should be appreciated that the main idea of the PCA dimension reduction algorithm is to map n-dimensional features onto k-dimensions (k, n being a natural number, k being smaller than n), which are completely new orthogonal features, also called principal components, are k-dimensional features reconstructed on the basis of the original n-dimensional features. Specifically, the method of dimension reduction processing includes, for example: the computing device 110 centers an input sample set for the weighted features, the input sample set having a first number of dimensions; calculating a covariance matrix of the input sample set; performing eigenvalue decomposition on the calculated covariance matrix so as to take out eigenvectors corresponding to the largest second number of eigenvalues, thereby normalizing the eigenvectors to form an eigenvector matrix, wherein the second number is smaller than the first number; and transforming each sample in the input sample set with the feature vector matrix to output an output sample set for the reduced dimension data, the output sample set having a second number of dimensions.

For example, the input of the dimension-reduction algorithm is, for example, the output of the n-dimensional sample set dimension-reduction algorithm with respect to the data subjected to the normalization processing is, for example, the sample set after dimension reduction. Where m represents the number of samples. The output of the dimension reduction algorithm is, for example, the sample set after dimension reduction +.>. Its dimension is for example +.>。/>Less than n. In one placeIn some embodiments, n is, for example and without limitation, 80./>For example, but not limited to, 4 or 5.

First, the computing device 110 targets an input sample set, i.e., an n-dimensional sample set, with respect to data subjected to normalization processing, for example, based on an algorithm exemplified by the following equation (13)Is centered.

（13）

In the above-mentioned formula (13),representing the ith sample in the n-dimensional input sample set D for data via normalization processing. j represents a variable. m represents the total number of samples.

The computing device 110 then calculates a covariance matrix of the input sample set D.

The computing device 110 then performs eigenvalue decomposition on the calculated covariance matrix, in order to extract a maximum second number (e.g.,a second number of eigenvectors corresponding to the eigenvalues so as to normalize all eigenvectors to form an eigenvector matrix W>Less than the first number n. It will be appreciated that the sample +. >Is->The principal component of the dimension is essentially the front +_of the covariance matrix of the input sample set>And characteristic values.

Furthermore, computing device 110 utilizes a feature vector matrix for each sample in the input sample setTransforming to generate an output sample set of reduced dimension data>. The algorithm of the above transformation is exemplified below in conjunction with equation (14):

（14）

in the above-mentioned formula (14),representing the output reduced-dimension data (i.e. output sample set +.>The ith sample) of (b). W represents a feature vector matrix. />Representing a transpose of the feature vector matrix. />Representing the ith sample in the n-dimensional input sample set D.

With respect to determining output sample setsDimension->Method of (i.e. dimension of reduced dimension data), in some embodiments,/->May be a preset value. At the position ofIn other embodiments, the dimension of the reduced dimension data is determined by a gravity threshold t that indicates the reduced dimension principal component. The specific gravity threshold t is set at (0, 1)]Between them. The algorithm for calculating the dimension of the reduced-dimension data is exemplarily described below in conjunction with equation (15).

（15）

In the above formula (15), t represents a specific gravity threshold value. n represents the characteristic value before dimension reductionIs a number of (3). />Representing the characteristic value after dimension reduction +. >Is a number of (3). />Representing characteristic value->。

At step 604, computing device 110 generates a real-time vector corresponding to the current operating condition based on the reduced-dimension data.

By adopting the means, the invention can construct the matched real-time vector which reflects the importance degree of the feature, retains the main components of the feature and has low latitude aiming at the detection data of different working conditions.

Fig. 7 illustrates a flowchart of a method 700 of generating a reference vector predictive model in accordance with an embodiment of the invention. It should be appreciated that the method 700 may be performed, for example, at the electronic device 1000 depicted in fig. 10. May also be performed at the computing device 110 depicted in fig. 1. It should be appreciated that method 700 may also include additional actions not shown and/or may omit actions shown, the scope of the present invention being not limited in this respect.

At step 702, computing device 110 clusters real-time vectors for each of the different operating conditions to determine a reference vector corresponding to each operating condition based on a cluster center. For example, the computing device 110 employs a clustering algorithm to cluster the real-time vectors under each working condition, for example, the cluster center is set to be 1, and the cluster center is the reference vector of the working condition. In some embodiments, the clustering algorithm is, for example and without limitation, the Kmeans method.

Specifically, regarding the method of clustering real-time vectors for each of different conditions, it includes, for example: the computing device 110 randomly selects K samples from the data set for the real-time vector for each of the different conditions as the initial cluster centerThe method comprises the steps of carrying out a first treatment on the surface of the For each sample in the dataset +.>Calculate per sample +.>Distance to K initial cluster centers in order to +.>Dividing the clustering result into categories corresponding to initial clustering centers with the smallest distance; recalculating the cluster center of each class +.>(the recalculated cluster center is the centroid of all samples of the class, +.in the above formula)>Representing a set of samples divided into such columns. />Representing any one of the samples in the sample set. />Representative sample set +.>Length of (d) a); repeat every sample +.>A step of sorting to the category having the smallest cluster center from the recalculation, and repeating the step of recalculating the cluster center until the position of the cluster center is no longer changed; and determining a reference vector corresponding to each operating condition based on the cluster center that is no longer changing.

At step 704, computing device 110 caches the determined reference vectors corresponding to each operating condition to a reference vector library.

For example, the reference vector corresponding to each working condition and the corresponding working condition (the working condition includes the characteristics such as temperature, rotating speed and flow) are cached in a reference vector library. For example, taking the rotational speed characteristics of the operating mode data as an example, for an operating mode with a rotational speed of 3000 revolutions, its corresponding reference vector indicates, for example, centroid coordinate data (25, 30). Similarly, different rotational speed characteristics may be mapped to corresponding centroid coordinate data to form a reference vector corresponding to each operating condition.

At step 706, computing device 110 generates a reference vector predictive model using a multiple linear regression model based on the condition data cached by the reference vector library, the reference vector corresponding to the condition.

As for the reference vector prediction model, it is constructed based on a multiple linear regression model, for example. The input data of the reference vector predictive model is, for example, operating mode data. The output data of the reference vector prediction model is, for example, a prediction reference vector corresponding to the working condition data.

The following illustrates a prediction function expression of a reference vector prediction model constructed based on a multiple linear regression model in conjunction with equations (16) and (17), which indicates an expression for predicting a certain dimension of a reference vector from the characteristics of the operating condition x. It should be appreciated that multiple models may be constructed to predict each dimension of the reference vector separately to determine the entire reference vector. In some embodiments, the reference vector comprises, for example, 4 or 5 dimensions. The following equation (18) illustrates the multi-function loss function expression of the reference vector prediction model for a certain dimension:

（16）

（17）

（18）

In the above formulas (16) to (18), m represents the total number of samples with respect to the operating condition and the reference vector. k represents the kth sample. n represents the total feature number for the operating mode. i represents the ith feature. X is x _i Representing the ith feature of the operating conditions x, each operating condition x, for example, comprising n features. In some embodiments, the operating conditions include characteristics such as temperature, rotational speed, flow rate, etc.I.e.,representing the corresponding weights of the n features of the operating condition x. />I.e. ] a +>Representing a loss function. />Representing a predicted value of a dimension of the reference vector corresponding to the condition x. />Representing a predicted value for a dimension of the reference vector corresponding to the condition x for the kth sample. />Representing the true value of a dimension of the reference vector corresponding to the condition x for the kth sample.

For example, solving the loss function using gradient descent methodFor any one weight->The bias derivative is calculated to obtain an expression shown in the following formula (19):

（19）

the parameter update equation for the reference vector predictive model is illustrated below in conjunction with equation (20):

（20）

in the above formulas (19) to (20),representing the learning rate. />Representing the corresponding weight of the jth feature. />The AND weight representing the kth sample +. >Corresponding->. m represents the total number of samples for the operating mode and reference vector. />Representing the operating condition x of the kth sample. />Representing a predicted value for a dimension of the reference vector corresponding to the condition x for the kth sample. />Representing the true value of a dimension of the reference vector corresponding to the condition x for the kth sample.

It should be appreciated that by the above method, the corresponding weights of the features in the operating condition data may be determined via a reference vector predictive model trained on samples of the operating condition and reference vectors.

The reference vector prediction model is then saved. Thus, the working condition data can be used as an independent variable, each dimension value of the reference vector corresponding to the working condition is used as an independent variable, and the reference vector corresponding to the working condition is predicted by adopting the reference vector prediction model constructed based on the multiple linear regression model, so as to obtain the predicted reference vector corresponding to the working condition.

By adopting the means, the method and the device can accurately predict the matched target reference vector even if the working condition corresponding to the real-time vector does not exist in the reference vector library.

Fig. 8 schematically illustrates a block diagram of a system 800 implementing a method for predicting the remaining useful life of a rotating equipment in accordance with further embodiments of the present invention. The system 800 includes, for example, a feature data preprocessing module 810, a target reference vector prediction module 820, and a health status data calculation module 830, a remaining useful life calculation module 840.

The feature data preprocessing module 810 is mainly used for data acquisition, data quality monitoring and feature extraction. Specifically, the feature data preprocessing module 810 is configured to: acquiring acceleration detection data about each detection position of the rotating equipment and working condition data about the working condition of the rotating equipment; determining whether the acquired acceleration detection data is abnormal; if it is determined that the acquired acceleration detection data is not abnormal, extracting features for the acceleration detection data so as to generate time domain features and frequency domain features.

With respect to the target reference vector prediction module 820, it is mainly used for real-time vector, vector clustering, and reference vector library caching. Specifically, the target reference vector prediction module 820 is configured to: clustering is carried out on the real-time vectors under each working condition in different working conditions so as to determine a reference vector corresponding to each working condition based on a clustering center; and generating a reference vector prediction model by using a multiple linear regression model based on the working condition data cached by the reference vector library and the reference vector corresponding to the working condition.

The health status data calculation module 830 is mainly used for initial health status data calculation, target reference vector library judgment, real-time vector calculation, health status data generation and health status data noise reduction. Specifically, the target reference vector prediction module 830 is configured to: determining initial health status data of the pump based on current vibration intensity value data of the pump; performing standardization and dimension reduction processing on time domain features and frequency domain features of working condition data of the current working condition so as to enable the variance contribution rate of the dimension reduced data to be larger than a preset variance threshold; based on the data after dimension reduction, generating a real-time vector corresponding to the current working condition; confirming whether a reference vector of a working condition corresponding to the real-time vector exists in a reference vector library; in response to determining that no reference vector exists in the reference vector library for the operating condition corresponding to the real-time vector, determining a predicted reference vector generated based on the reference vector prediction model as a target reference vector; and in response to determining that the reference vector library has the reference vector of the working condition corresponding to the real-time vector, determining the reference vector of the working condition corresponding to the real-time vector as a target reference vector; calculating the distance between the real-time vector and the target reference vector; and generating health status data about the rotating equipment based on the calculated distance, correcting the calculated target health status data about the rotating equipment based on the determined initial health status data to obtain corrected target health status data; and denoising the health degree indicating data by using a denoising function to acquire target health state evaluation data about the rotating equipment.

The remaining service life calculation module 840 is mainly used for data caching, inflection point judgment, target machine model selection, RUL model fitting, target machine model parameter updating, model fitting effect judgment, and remaining service life calculation. Specifically, the target reference vector prediction module 840 is configured to: caching target health status data about the rotating equipment generated within a predetermined period; and in response to determining that the health status trend curve exhibits an inflection point, determining the cached target health status data after the occurrence of the inflection point as health decay data for the rotating device; determining whether an inflection point appears in a historical health state data curve formed based on the target health state data; responsive to determining that the health status trend curve exhibits inflection points, re-fitting the historical health status data curve with the plurality of candidate machine learning models to select a matched target machine learning model among the plurality of candidate machine learning models; training the determined target machine learning model based on the health decay data to update parameters of the target machine learning model; fitting the historical health state data curve by using the updated parameter target machine learning model so as to determine whether the fitting effect meets a preset condition; generating a residual service life prediction curve by adopting the fitted target machine learning model in response to determining that the fitting effect meets a preset condition; determining an intersection point of the generated residual service life prediction curve and a specific health degree; and calculating the time length between the current time and the time corresponding to the intersection point so as to determine the residual service life of the rotating equipment based on the time length.

Fig. 10 schematically shows a block diagram of an electronic device 1000 suitable for use in implementing embodiments of the invention. The electronic device 1000 may be for implementing the methods 200-700 shown in fig. 2-7. As shown in fig. 5, the electronic device 1000 includes a central processing unit (i.e., a CPU 1001) that can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (i.e., a ROM 1002) or computer program instructions loaded from a storage unit 1008 into a random access memory (i.e., a RAM 1003). In the RAM 1003, various programs and data required for the operation of the electronic apparatus 1000 can also be stored. The CPU 1001, ROM 1002, and RAM 1003 are connected to each other by a bus 1004. An input/output interface (i.e., I/O interface 1005) is also connected to bus 1004.

Various components in the electronic device 1000 are connected to the I/O interface 1005, including: the cpu 1001 performs the respective methods and processes described above, for example, performs the methods 200 to 700, the input unit 1006, the output unit 1007, and the storage unit 1008. For example, in some embodiments, the methods 200-700 may be implemented as a computer software program stored on a machine-readable medium, such as the storage unit 1008. In some embodiments, part or all of the computer program may be loaded and/or installed onto the electronic device 1000 via the ROM 1002 and/or the communication unit 1009. When the computer program is loaded into RAM 1003 and executed by CPU 1001, one or more operations of methods 200 through 700 described above may be performed. Alternatively, in other embodiments, CPU 1001 may be configured to perform one or more actions of methods 200-700 in any other suitable manner (e.g., by way of firmware).

It should be further appreciated that the present invention can be a method, apparatus, system, and/or computer program product. The computer program product may include a computer readable storage medium having computer readable program instructions embodied thereon for performing various aspects of the present invention.

The computer readable storage medium may be a tangible device that can hold and store instructions for use by an instruction execution device. The computer readable storage medium may be, for example, but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium would include the following: portable computer disks, hard disks, random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static Random Access Memory (SRAM), portable compact disk read-only memory (CD-ROM), digital Versatile Disks (DVD), memory sticks, floppy disks, mechanical coding devices, punch cards or in-groove structures such as punch cards or grooves having instructions stored thereon, and any suitable combination of the foregoing. Computer-readable storage media, as used herein, are not to be construed as transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., optical pulses through fiber optic cables), or electrical signals transmitted through wires.

The computer readable program instructions described herein may be downloaded from a computer readable storage medium to a respective computing/processing device or to an external computer or external storage device over a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmissions, wireless transmissions, routers, firewalls, switches, gateway computers and/or edge servers. The network interface card or network interface in each computing/processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium in the respective computing/processing device.

Computer program instructions for carrying out operations of the present invention may be assembly instructions, instruction Set Architecture (ISA) instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, c++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer readable program instructions may be executed entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computer (for example, through the Internet using an Internet service provider). In some embodiments, aspects of the present invention are implemented by personalizing electronic circuitry, such as programmable logic circuitry, field Programmable Gate Arrays (FPGAs), or Programmable Logic Arrays (PLAs), with state information for computer readable program instructions, which can execute the computer readable program instructions.

Various aspects of the present invention are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer-readable program instructions.

These computer readable program instructions may be provided to a processor in a voice interaction device, a processing unit of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processing unit of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer readable medium having the instructions stored therein includes an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks.

The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer, other programmable apparatus or other devices implement the functions/acts specified in the flowchart and/or block diagram block or blocks.

The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of devices, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The foregoing description of embodiments of the invention has been presented for purposes of illustration and description, and is not intended to be exhaustive or limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the various embodiments described. The terminology used herein was chosen in order to best explain the principles of the embodiments, the practical application, or the technical improvements in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

The above is only an alternative embodiment of the present invention and is not intended to limit the present invention, and various modifications and variations will be apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting the remaining useful life of a rotating equipment, comprising:

acquiring acceleration detection data about each detection position of the rotating equipment and working condition data about the working condition of the rotating equipment;

Preprocessing the acquired acceleration detection data so as to generate time domain features and frequency domain features;

generating a real-time vector corresponding to the current operating condition based on the generated time domain features and frequency domain features so as to calculate target health state data about the rotating equipment based on the real-time vector and the target reference vector;

determining whether an inflection point appears in a historical health status data curve formed based on the cached target health status data; and

responsive to determining that the historical state of health data curve exhibits inflection points, re-fitting the historical state of health data curve with the plurality of candidate machine learning models to select a matched target machine learning model from the plurality of candidate machine learning models; training the determined target machine learning model based on the health decay data to update parameters of the target machine learning model; fitting the historical health state data curve by using the updated parameter target machine learning model so as to determine whether the fitting effect meets a preset condition; and in response to determining that the fitting effect meets a predetermined condition, predicting the remaining service life of the rotating equipment by adopting a fitted target machine learning model, wherein the health decay data are target health state data cached after the inflection point of the historical health state data curve appears.

2. The method as recited in claim 1, further comprising:

the target health status data about the rotating equipment generated within a predetermined period is buffered, the predetermined period being adjustable.

3. The method of claim 1 or 2, wherein determining whether an inflection point appears in a historical health data curve formed based on the cached target health data comprises:

generating a historical health status data curve based on the cached target health status data;

dividing the historical health status data curve into a plurality of sliding windows;

calculating an inclination angle of each sliding window in the plurality of sliding windows; and

in response to determining that the tilt angle of the current sliding window exceeds the predetermined tilt angle threshold range, it is determined that an inflection point occurs at the current sliding window.

4. The method of claim 1, wherein predicting a remaining useful life of the rotating equipment using the fitted target machine learning model comprises:

generating a residual service life prediction curve by adopting the fitted target machine learning model;

determining an intersection point of the generated remaining life prediction curve and the specific health state data; and

And calculating the time length between the current time and the time corresponding to the intersection point so as to determine the residual service life of the rotating equipment based on the time length.

5. The method of claim 4, wherein the remaining life prediction curve is fitted using an exponential model whose parameters are calculated based on a least squares method.

6. The method of claim 1, wherein the rotating apparatus is a pump, the method further comprising:

based on the current vibration intensity value data of the pump, initial health state data of the pump is determined.

7. The method of claim 6, wherein determining initial health status data for the pump comprises any one of:

determining a vibration level of the pump based on the current vibration intensity value data of the pump, so as to determine initial health state data of the pump based on the vibration level of the pump; or alternatively

Initial health status data of the pump is determined based on the current vibration intensity value data of the pump and a predetermined mapping algorithm.

8. The method of claim 1, wherein generating a real-time vector corresponding to a current operating condition based on the generated time-domain features, frequency-domain features comprises:

Performing standardization and dimension reduction processing on time domain features and frequency domain features of working condition data of the current working condition so as to enable the variance contribution rate of the dimension reduced data to be larger than a preset variance threshold; and

and generating a real-time vector corresponding to the current working condition based on the dimension-reduced data.

9. The method of claim 6, wherein calculating target health status data for a rotating device comprises:

confirming whether a reference vector of a working condition corresponding to the real-time vector exists in a reference vector library;

in response to determining that no reference vector exists in the reference vector library for the operating condition corresponding to the real-time vector, determining a predicted reference vector generated based on a reference vector prediction model as a target reference vector matching the real-time vector;

in response to determining that a reference vector of a working condition corresponding to a real-time vector exists in a reference vector library, determining the reference vector of the working condition corresponding to the real-time vector as a target reference vector matched with the real-time vector;

calculating the distance between the real-time vector and the target reference vector; and

generating target health status data about the rotating device based on the calculated distance;

based on the determined initial health state data, a correction is made for the calculated target health state data about the rotating equipment to obtain corrected target health state data.

10. The method of claim 9, wherein the reference vector prediction model is generated via:

clustering is carried out on the real-time vectors under each working condition in different working conditions so as to determine a reference vector corresponding to each working condition based on a clustering center;

caching the determined reference vector corresponding to each working condition to a reference vector library, and

and generating a reference vector prediction model by utilizing the multiple linear regression model based on the working condition data cached by the reference vector library and the reference vector corresponding to the working condition.

11. The method of claim 1, wherein the acceleration detection data comprises: acceleration detection data of the pump driving end bearing in the vertical direction, acceleration detection data of the pump driving end bearing in the horizontal direction and acceleration detection data of the pump driving end bearing in the axial direction, wherein the working condition data is at least one of temperature, rotating speed and flow; the time domain features include: regarding a plurality of effective values, peaks, kurtosis coefficients of the acceleration detection data, the frequency domain features include at least one of: the inner ring fault frequency energy, the outer ring fault frequency energy, and the pump drive end and non-drive end vertical direction phase differences of the pump are related.

12. A computing device, comprising:

at least one processing unit;

at least one memory coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, which when executed by the at least one processing unit, cause the apparatus to perform the steps of the method according to any one of claims 1 to 11.

13. A computer readable storage medium having stored thereon a computer program which, when executed by a machine, implements the method according to any of claims 1 to 11.