Method for generating motor life label dataset based on discrete exponential degradation model

Abstract

The present application relates to the field of label data set generation, and specifically discloses a motor life label data set generation method based on a discrete exponential degradation model, comprising the following steps: S1, extracting the steady-state current of each action; S2, calculating the difference between the steady-state currents in each action to obtain a degradation feature; S3, determining a degradation control limit based on the statistical characteristics of the degradation feature in a preset health period, comparing the degradation feature after the preset health period with the degradation control limit, determining a first degradation time point and obtaining an actual health period; S4, representing the degradation degree of all current actions based on the degradation feature in the actual health period; S5, constructing a discrete exponential degradation recursive model in the action frequency domain based on the degradation degree; S6, analytically solving the residual life label corresponding to each action based on the discrete exponential degradation recursive model; and S7, forming a data set according to the residual life label. The present application adapts to individual differences between different devices and has high label accuracy.

Description

Technical Field

[0001] This application relates to the field of tag dataset generation, specifically to a method for generating motor life tag datasets based on a discrete exponential degradation model. Background Technology

[0002] Predictive maintenance, by identifying and warning of equipment degradation in advance, can avoid sudden failures and reduce operation and maintenance costs, making it a core method of current industrial equipment operation and maintenance management. Remaining life (RUL) n i n The "Useful Life" label and the training dataset labeled with "Remaining Life" are key inputs for training the remaining life prediction model and building a data-driven operation and maintenance system.

[0003] Electrical equipment (such as platform screen doors, automatic door operators, servo motors, and electric actuators) exhibits significant nonlinear characteristics in its degradation process during long-term operation. For example, wear on motor bearings leads to an accelerated increase in the rate of change of current as degradation intensifies, showing a "the more it wears, the faster it gets" accelerated degradation trend. Furthermore, directly collected raw motor current data contains a large amount of pulse width modulation chopper noise and transient impacts during start-stop operations, making it difficult to directly characterize the degradation state. Therefore, accurately extracting degradation features from raw motor current data and generating standardized datasets labeled with remaining life is a crucial preliminary step in predicting the remaining life of electrical equipment.

[0004] Currently, the existing technical solutions for generating remaining life labels for electric equipment mainly include the following categories: Firstly, there is the remaining life label generation method based on linear models. This type of method uses linear regression or other linear models to fit the trend of degradation characteristics changing with the number of operations, and then extrapolates to solve for the remaining life. Since linear models assume that the degradation rate is a constant value, they cannot capture the accelerated degradation law of electric equipment, which is "faster with wear," and the prediction error increases significantly in the accelerated degradation stage.

[0005] Secondly, there is the remaining lifetime label generation method based on the continuous exponential model. This type of method describes the degradation process using continuous differential equations and directly applies the analytical solution of the continuous model to calculate the remaining lifetime. However, the number of actions of electrical equipment is a discrete variable. Directly applying the analytical solution of the continuous model to a discrete action sequence results in fixed model parameters, making it unable to adapt to the individual degradation trajectories of different devices.

[0006] Third, there is the method for determining the first degradation and generating the remaining lifetime tag based on manually set thresholds. This method involves manually setting a current threshold based on experience to determine the time point of the first degradation and generating the remaining lifetime tag accordingly. Manual threshold setting is highly subjective, and the method of determining degradation based on a single point exceeding the limit lacks an anti-interference mechanism, making it susceptible to transient spikes and false triggering, resulting in low tag accuracy. Summary of the Invention

[0007] In view of this, the present invention proposes a method for generating motor life label datasets based on a discrete exponential degradation model, comprising the following steps: S1, acquire motor current data of electric equipment during each operation, segment the motor current data, and extract the steady-state current of each operation. Each operation includes a pair of adjacent opening and closing operations. S2, calculate the difference in steady-state current during each action to obtain the degradation characteristics; S3, determine the degradation control limit based on the statistical characteristics of the degradation features within the preset health period, compare the degradation features after the preset health period with the degradation control limit to determine the first degradation time point, and merge the actions before the first degradation time point into the preset health period to obtain the actual health period; S4, based on the degenerative characteristics during the actual health period, characterize the degree of degeneration of all current actions; S5. Based on the degree of degradation, construct a discrete exponential degradation recursive model in the action count domain to characterize the degree of degradation of future actions; S6. Based on the discrete exponential degradation recursive model, the remaining lifetime label corresponding to each action is analytically solved. S7, A dataset is formed based on the remaining lifespan labels.

[0008] Furthermore, in S1, The motor current data for each action is filtered and denoised. The steady-state segment for each action is determined by the sliding window variance threshold method. The mean value of the current data within the steady-state segment is taken as the steady-state current. The steady-state current includes the steady-state current during the start-up and stop-down operations.

[0009] Furthermore, in S2, The degradation characteristic is the difference between the steady-state current of the shutdown action and the steady-state current of the startup action within each operation. The formula for calculating the degradation characteristic is as follows: , Where, Δ I ( k ) is the first k Degenerative characteristics of secondary actions. I close( k ) is the first k The steady-state current of the shut-off action described in this action. I open ( k ) is the first k The steady-state current for starting the action described in this action.

[0010] Furthermore, in S3, The initial M actions are selected as the preset health period. The mean μ and standard deviation σ of the degradation characteristics within the preset health period are calculated, and the degradation control limit is determined as UCL = μ + 3σ. Starting from the (M+1)th action, when the degradation feature continuously crosses the degradation control limit a preset number of times, the sequence number of the action that first crosses the degradation control limit is determined as the first degradation time point, and the actions before the first degradation time point are merged into the preset health period to obtain the actual health period.

[0011] Furthermore, in S4, The degree of degradation is characterized by a degradation index, which is calculated using the following formula: , Among them, DI ( k ) is the first k The degradation index of the sub-action, Δ I ( k ) is the first k Degenerative characteristics of secondary actions, Δ I healthy Δ is the mean of the degenerative characteristics during the actual healthy period. I fail The preset failure degradation characteristic threshold, DI( k The higher the DI, the more severe the degradation. k )=0 indicates perfect health, DI( k =1 indicates complete degradation.

[0012] Furthermore, in S5, The discrete exponential degeneracy recursive model is DI( K +1) = (1 + β )·DI( K ) + l ;in, K DI( is the number of times all actions are performed.) K +1) is for the first K The degradation index of +1 action prediction, DI( K ) is the first K Degeneration index of the sub-action. l For inherent degradation increment, β For acceleration coefficient; when β When the value equals 0, the discrete exponential degeneracy recursive model is linearly degenerate.

[0013] Furthermore, S5 also includes: The inherent degradation increment is calculated using a recursive least squares method with a forgetting factor. l and the acceleration coefficient β The forgetting factor is updated online and its value range is (0.9, 0.99).

[0014] Furthermore, S5 also includes: Based on the aforementioned discrete exponential degeneracy recursive model, for the future... n The degradation index of the sub-action is used to predict: when β When greater than 0: ; when β When equal to 0: ; Among them, DI ( K + n ) for the first K + n Degradation index predicted by sub-action.

[0015] Furthermore, in S6: Let DI( K + n =1, solve. n This refers to the remaining lifespan label. when β When the value is greater than 0, the formula for calculating the remaining lifetime tag is: ; when β When the value is 0, the formula for calculating the remaining lifetime tag is: ; Among them, RUL ( k ) is the first k The remaining lifetime tag for this action.

[0016] Furthermore, in step S7, the dataset, indexed by action sequence number, also includes motor current data, steady-state current, degradation characteristics, and relevant parameters of the discrete exponential degradation model. This invention uses statistical characteristics to determine the degradation control limits, avoiding the subjectivity of manually setting thresholds. The control limits can be automatically determined based on the initial state of the equipment, thus adapting to individual differences between different devices. By screening degradation features within the actual healthy period, the degradation degree of all current actions is characterized. A discrete exponential degradation recursive model is constructed in the action frequency domain, thereby capturing the accelerated degradation law of electric equipment as it wears down faster. The labeling accuracy is high and can be directly used for training downstream prediction models. Attached Figure Description

[0017] Figure 1 This is a schematic diagram illustrating the steps of the method for generating motor life tag datasets based on the discrete exponential degradation model of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the various embodiments of this application will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details have been presented in the various embodiments of this application to enable the reader to better understand this application. However, the technical solutions claimed in this application can be implemented even without these technical details and various changes and modifications based on the following embodiments.

[0019] In the following description, certain specific details are set forth for the purpose of illustrating various disclosed embodiments in order to provide a thorough understanding of the various disclosed embodiments. However, those skilled in the art will recognize that embodiments may be practiced without one or more of these specific details. In other instances, well-known apparatuses, structures, and techniques associated with this application may not have been shown or described in detail to avoid unnecessarily obscuring the description of the embodiments.

[0020] Unless the context requires otherwise, throughout the specification and claims, the word “comprising” and its variations, such as “including” and “having”, shall be understood to have an open, inclusive meaning, that is, to be interpreted as “including, but not limited to”.

[0021] Throughout this specification, references to "an embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Therefore, the appearance of "in an embodiment" or "an embodiment" in various places throughout the specification does not necessarily refer to the same embodiment. Furthermore, a particular feature, structure, or characteristic may be combined in any manner in one or more embodiments.

[0022] The purpose of this invention is to accurately extract degradation features from raw motor current data and generate a standardized dataset with remaining life labels. Specifically, it proposes a method for generating motor life-labeled datasets based on a discrete exponential degradation model, such as... Figure 1As shown, it includes the following steps: S1, acquire motor current data of electric equipment during each operation, segment the motor current data, and extract the steady-state current of each operation. Each operation includes a pair of adjacent opening and closing operations. S2, calculate the difference in steady-state current during each action to obtain the degradation characteristics; S3, determine the degradation control limit based on the statistical characteristics of the degradation features within the preset health period, compare the degradation features after the preset health period with the degradation control limit to determine the first degradation time point, and merge the actions before the first degradation time point into the preset health period to obtain the actual health period; S4, based on the degenerative characteristics during the actual health period, characterize the degree of degeneration of all current actions; S5. Based on the degree of degradation, construct a discrete exponential degradation recursive model in the action count domain to characterize the degree of degradation of future actions; S6. Based on the discrete exponential degradation recursive model, the remaining lifetime label corresponding to each action is analytically solved. S7, A dataset is formed based on the remaining lifespan labels.

[0023] In S1, The motor current data for each action is filtered and denoised. The steady-state segment for each action is determined by the sliding window variance threshold method. The mean value of the current data within the steady-state segment is taken as the steady-state current. The sliding window variance threshold method determines the steady-state segment for each action. That is, the variance of the current signal within the sliding window is calculated. When the variance is continuously lower than the set threshold, the interval is determined to have entered the steady-state segment, and the mean value of the current data within the steady-state segment is taken as the steady-state current for that action.

[0024] The steady-state current includes the steady-state current during the start-up and stop-down operations.

[0025] The motor current waveform of electric equipment during operation typically includes start-stop transient impacts and a steady-state operation phase. Start-stop transient impacts are sudden current changes occurring at the instant the motor starts and stops; they are highly random and volatile, and cannot stably reflect the degradation state of the equipment. In contrast, the current level during the steady-state operation phase is closely related to the degree of equipment degradation and exhibits better stability. Therefore, this step extracts the steady-state current from the motor current as a fundamental characteristic characterizing the equipment's degradation state. The steady-state current refers to the average current during the steady-state operation phase after excluding start-stop transient disturbances.

[0026] In S2, the degradation characteristic is the difference between the steady-state current of the shut-off action and the steady-state current of the turn-on action within each action. The formula for calculating the degradation characteristic is: , Where, Δ I ( k ) is the first k Degenerative characteristics of secondary actions. I close ( k ) is the first k The steady-state current of the shut-off action described in this action. I open ( k ) is the first k The steady-state current for starting the action described in this action.

[0027] Since the turning-on and turning-off actions occur under the same operating environment and conditions, they are subject to essentially the same common-mode interference, such as temperature drift and power supply fluctuations. By subtracting the steady-state current of the turning-off action from that of the turning-on action, the effects of the aforementioned common-mode interference can be offset, thereby extracting degradation characteristics that reflect the essence of equipment degradation.

[0028] Furthermore, in S3, the initial first M actions are selected as the preset health period, the mean μ and standard deviation σ of the degradation characteristics within the preset health period are calculated, and the degradation control limit is determined to be UCL=μ+3σ. The preset health period refers to the initial operating range used to estimate the statistical characteristics of the equipment's health status, while the degradation control limit (UCL) refers to the critical value used to determine whether degradation has occurred.

[0029] Starting from the (M+1)th action, when the degradation feature continuously crosses the degradation control limit a preset number, such as 3 times, the sequence number of the action that first crosses the degradation control limit is determined as the first degradation time point, and the actions before the first degradation time point are merged into the preset health period to obtain the actual health period.

[0030] Determining degradation control limits using statistical methods, rather than manually setting thresholds, avoids the subjectivity of manually setting thresholds and allows control limits to be automatically determined based on the initial state of the equipment, thus adapting to individual differences between different devices.

[0031] Furthermore, in S4, The degree of degradation is characterized by a degradation index, which is calculated using the following formula: , Among them, DI ( k ) is the first k The degradation index of the sub-action, Δ I ( k ) is the first k Degenerative characteristics of secondary actions, Δ I healthyΔ is the mean of the degenerative characteristics during the actual healthy period. I fail The preset failure degradation characteristic threshold is calibrated by statistical analysis of historical current differences when the equipment experiences severe faults such as jamming. If actual failure data is lacking, it can be calibrated according to the equipment design specifications or accelerated life tests, such as taking the 90% confidence lower limit of the critical current difference of the equipment jamming.

[0032] DI( k The higher the DI, the more severe the degradation. k )=0 indicates perfect health, DI( k )=1 indicates complete degradation. DI( K ) is the first K The degradation index of the current action is the degradation index of the last action.

[0033] In S5, The discrete exponential degeneracy recursive model is DI( K +1)=(1+ β )·DI( K )+ l ;in, K DI( is the number of times all actions are performed.) K +1) is for the first K The degradation index of +1 action prediction, DI( K ) is the first K Degeneration index of the sub-action. l For inherent degradation increment, β For acceleration coefficient; when β When the value equals 0, the discrete exponential degeneracy recursive model is linearly degenerate.

[0034] In the above, l It reflects the basic degradation amount per operation of the equipment. β It reflects the amplification effect of the current degree of equipment degradation on the future degradation rate. The degradation increment of the equipment at the next moment is the sum of the inherent degradation increment and the accelerated degradation increment, accurately depicting the nonlinear law that wear gets faster with wear.

[0035] To enable the model to adapt to individual degradation differences across different devices, this step also involves online updates to the inherent degradation increment and acceleration coefficient. In one embodiment, a recursive least squares method with a forgetting factor is used to update the model. l and β Online updates are performed, with the forgetting factor ranging from (0.9 to 0.99). The forgetting factor assigns different weights to old and new data, enabling the model to quickly track changes in the device's current degradation trend. The computational complexity of a single update using the recursive least squares method is constant, meeting the real-time requirements of high-frequency operation scenarios.

[0036] In the initial stage of equipment commissioning, data from the first few operations can be used to... l and β Offline fitting is performed to obtain initial parameter values. The fitting results vary for different electric devices (such as platform screen doors, automatic door operators, servo motors, electric actuators, etc.) to reflect the characteristics of different electric devices. During the recursive update process, if the parameter estimate is negative, it is judged as invalid and clamped to zero to ensure that the physical meaning of the model is not violated.

[0037] Based on the above recursive model, we can also predict the future. n The degradation index of the sub-action is predicted. When β When greater than 0, the first K + n The degradation index of this action satisfies: ; when β When equal to 0, the first K + n The degradation index of this action satisfies: ; Among them, DI ( K + n ) for the first K + n Degradation index for sub-action prediction. n This represents the number of steps in the extrapolation. The analytical solution is rigorously derived from the recursive model, and there is no issue of dimensional inconsistency caused by discretization of a continuous model.

[0038] Step S6: Based on the discrete exponential degradation recursive model, the remaining lifetime label corresponding to each action is analytically solved.

[0039] The remaining lifetime label refers to the number of actions a device has remaining between its current number of actions and its eventual failure. When solving for this label, it is assumed that the parameter estimates at the current moment will remain constant in the future, and that the degradation index (DI) reaches the failure state at a future action. K + n )=1, and based on this, the remaining number of actions can be solved.

[0040] when β When the value is greater than 0, the formula for calculating the remaining lifetime label is: ; when β When the remaining lifetime tag equals 0, the formula for calculating the remaining lifetime tag is: ; Among them, RUL ( k ) is the first kThe remaining lifetime tag for this action. Specifically, k for K The time is the remaining lifetime label of the last action. An analytical solution is used, avoiding errors caused by numerical iteration and ensuring the accuracy and reliability of the label.

[0041] Step S7: Generate a dataset based on the remaining lifespan labels.

[0042] Specifically, using the action sequence number as an index, the motor current data, steady-state current, degradation characteristics, degradation index, relevant parameters of the discrete exponential degradation recursive model, and remaining lifetime labels are aligned and concatenated according to the action dimension to form a structured dataset with remaining lifetime labels. Each sample in the dataset corresponds to one action and includes fields such as the original current segment, steady-state current, degradation characteristics, degradation index, model parameters, and remaining lifetime label for that action.

[0043] Furthermore, the dataset can be standardized, including at least one of the following operations: uniform sampling frequency, action alignment, missing value imputation, non-uniform interval action processing, and label normalization. This ensures that data from different devices and different collection batches have a uniform format and scale, thereby supporting cross-device model training. Subsequently, samples from multiple devices can be aggregated, divided into training, validation, and test sets according to device number, and a standardized dataset that can be directly used for training downstream lifetime prediction models can be output.

[0044] In scenarios where historical expert rating data exists, the remaining lifetime label generated by the physical model can be probabilistically fused with the expert experience rating based on statistical features to serve as the final label, thereby achieving complementary verification between physical modeling and expert knowledge.

[0045] The solutions provided by the embodiments of this application have been described in detail above. The various embodiments in the specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. It should be noted that those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

[0046] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0047] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

Claims

1. A method for generating motor life label datasets based on a discrete exponential degradation model, characterized in that, Includes the following steps: S1, acquire motor current data of electric equipment during each operation, segment the motor current data, and extract the steady-state current of each operation, wherein each operation includes a pair of adjacent opening and closing operations. S2, calculate the difference in steady-state current during each action to obtain the degradation characteristics; S3, determine the degradation control limit based on the statistical characteristics of the degradation features within the preset health period, compare the degradation features after the preset health period with the degradation control limit to determine the first degradation time point, and merge the actions before the first degradation time point into the preset health period to obtain the actual health period; S4, based on the degenerative characteristics during the actual health period, characterize the degree of degeneration of all current actions; S5. Based on the degree of degradation, construct a discrete exponential degradation recursive model in the action count domain to characterize the degree of degradation of future actions; S6. Based on the discrete exponential degradation recursive model, the remaining lifetime label corresponding to each action is analytically solved. S7, A dataset is formed based on the remaining lifespan labels.

2. The method for generating motor life tag dataset based on discrete exponential degradation model according to claim 1, characterized in that, In S1, The motor current data for each action is filtered and denoised. The steady-state segment for each action is determined by the sliding window variance threshold method. The mean value of the current data within the steady-state segment is taken as the steady-state current. The steady-state current includes the steady-state current during the start-up and stop-down operations.

3. The method for generating a motor life tag dataset based on a discrete exponential degradation model according to claim 2, characterized in that, In S2, The degradation characteristic is the difference between the steady-state current of the shutdown action and the steady-state current of the startup action within each operation. The formula for calculating the degradation characteristic is as follows: ， Where, Δ I ( k ) is the first k Degenerative characteristics of secondary actions. I close ( k ) is the first k The steady-state current of the shut-off action described in this action. I open ( k ) is the first k The steady-state current for starting the action described in this action.

4. The method for generating motor life tag dataset based on discrete exponential degradation model according to claim 1, characterized in that, In S3, The initial M actions are selected as the preset health period. The mean μ and standard deviation σ of the degradation characteristics within the preset health period are calculated, and the degradation control limit is determined as UCL = μ + 3σ. Starting from the (M+1)th action, when the degradation feature continuously crosses the degradation control limit a preset number of times, the sequence number of the action that first crosses the degradation control limit is determined as the first degradation time point, and the actions before the first degradation time point are merged into the preset health period to obtain the actual health period.

5. The method for generating a motor life tag dataset based on a discrete exponential degradation model according to claim 1, characterized in that, In S4, The degree of degradation is characterized by a degradation index, which is calculated using the following formula: ， Among them, DI ( k ) is the first k The degradation index of the sub-action, Δ I ( k ) is the first k Degenerative characteristics of secondary actions, Δ I healthy Δ is the mean of the degenerative characteristics during the actual healthy period. I fail The preset failure degradation characteristic threshold, DI( k The higher the DI, the more severe the degradation. k )=0 indicates perfect health, DI( k =1 indicates complete degradation.

6. The method for generating a motor life tag dataset based on a discrete exponential degradation model according to claim 5, characterized in that, In S5, The discrete exponential degeneracy recursive model is DI( K +1) = (1 + β )·DI( K ) + λ ; in, K DI( is the total number of actions performed so far.) K +1) is for the first K The degradation index of +1 action prediction, DI( K ) is the first K Degeneration index of the sub-action. λ For inherent degradation increment, β This is the acceleration coefficient; when β When the value equals 0, the discrete exponential degeneracy recursive model is linearly degenerate.

7. The method for generating a motor life tag dataset based on a discrete exponential degradation model according to claim 6, characterized in that, S5 also includes: The inherent degradation increment is calculated using a recursive least squares method with a forgetting factor. λ and the acceleration coefficient β The forgetting factor is updated online and its value range is (0.9, 0.99).

8. The method for generating a motor life tag dataset based on a discrete exponential degradation model according to claim 6, characterized in that, S5 also includes: Based on the aforementioned discrete exponential degeneracy recursive model, for the future... n The degradation index of the sub-action is used to predict: when β When greater than 0: ； when β When equal to 0: ； Among them, DI ( K + n ) for the first K + n Degradation index predicted by sub-action.

9. The method for generating a motor life tag dataset based on a discrete exponential degradation model according to claim 8, characterized in that, In S6: Let DI( K + n Given that ) = 1, solve for 1. n This refers to the remaining lifespan label. when β When the value is greater than 0, the formula for calculating the remaining lifetime tag is: ； when β When the value is 0, the formula for calculating the remaining lifetime tag is: ； Among them, RUL ( k ) is the first k The remaining lifetime tag for this action.

10. The method for generating a motor life tag dataset based on a discrete exponential degradation model according to claim 1, characterized in that, In step S7, the dataset is indexed by the action sequence number and also includes motor current data, steady-state current, degradation characteristics, and relevant parameters of the discrete exponential degradation model.

Images