Methods, devices and electronic equipment for predicting the remaining service life of mechanical equipment under time-varying operating conditions
By decoupling the health state model and the physical degradation state space model under operating conditions, the problem of decoupling degradation information and operating condition information in the prediction of the remaining service life of mechanical equipment under time-varying operating conditions is solved, and accurate prediction and stability improvement under complex operating conditions are achieved.
Patent Information
- Application Number
- CN202611175368.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-04
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies struggle to effectively decouple degradation information from operating condition information in predicting the remaining service life of mechanical equipment under time-varying operating conditions. This leads to the model learning pseudo-degradation features, and the long-term prediction accuracy and stability are insufficient, making it prone to error accumulation and state drift.
A decoupled health state model based on operating conditions and a physical degradation state space model are adopted. Degradation information and operating condition information are separated through signal coding network, operating condition coding network and health state coding network. The dynamic prediction and adaptive correction mechanism of the physical degradation state space model is used to achieve accurate characterization and long-term stable prediction of health state.
It enables accurate prediction of the remaining service life of mechanical equipment under complex and ever-changing time-varying operating conditions, reduces the interference of operating condition changes, suppresses error accumulation and state drift, and improves the stability and accuracy of prediction.
Smart Images

Figure CN123068831A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of mechanical equipment prediction and health management technology, specifically relating to a method, device and electronic equipment for predicting the remaining service life of mechanical equipment under time-varying operating conditions. Background Technology
[0002] Mechanical equipment is widely used in aerospace, energy and power, rail transportation, and intelligent manufacturing. The health status of its key components, such as rolling bearings, gears, and rotors, directly affects equipment operation safety and production efficiency. Predictive health management of mechanical equipment, accurately predicting its remaining service life, provides crucial information for equipment condition assessment, maintenance decisions, and predictive maintenance. This is significant for reducing maintenance costs, avoiding unexpected downtime, and ensuring safe equipment operation. Current mainstream data-driven remaining service life prediction methods primarily model under constant operating conditions. However, in actual industrial applications, equipment typically operates under constantly changing time-varying conditions, including varying speeds, loads, and temperatures. These changes in operating conditions alter the distribution of monitoring signals, resulting in signals that contain both equipment degradation information and interference from these changes.
[0003] Several attempts have been made in the prior art to detect equipment anomalies and predict remaining service life under time-varying operating conditions. For example, Chinese patent application CN121365338A discloses an integrated method and device for detecting anomalies and predicting remaining service life of rotating machinery under time-varying operating conditions. This method initializes a two-stage hybrid state-space model. In the degradation stage, the root mean square value of the vibration signal and the rotational speed at the current moment are input into the two-stage hybrid state-space model, and a nonlinear filtering algorithm is used to calculate the health state of the rotating machinery at the current moment. Subsequently, based on the health state at the current moment and before the current moment, the method selects the best degradation fitting model from multiple preset degradation fitting models based on model selection criteria, and updates the state transition coefficients based on the best degradation fitting model. Finally, based on the updated model, numerical simulation technology is used to predict the probability density distribution of the remaining service life.
[0004] Although the aforementioned existing technologies consider operating condition information such as rotational speed to some extent, they still have certain shortcomings in terms of practical and complex industrial applicability, mainly as follows: First, they only use the statistical characteristics of vibration signals and operating condition parameters such as rotational speed as direct inputs to the observation equations. This approach fails to effectively separate and decouple degradation information from operating condition information in the deep representation space, making it easy for the model to learn pseudo-degradation features related to changes in operating conditions, and making it difficult to accurately represent the true health state of the equipment. Second, the recursion of the state also relies on the discrete switching of several preset independent empirical degradation fitting models. The degradation of actual mechanical equipment is a physical process continuously modulated by time-varying operating conditions. Simply applying a discrete-time degradation fitting model makes it difficult to fully utilize the physical dynamic evolution law in the degradation process, resulting in the need to improve the long-term prediction accuracy and generalization ability of the model under complex time-varying operating conditions. Third, when performing state recursion, the state-space model uses nonlinear filtering algorithms such as extended Kalman filtering to calculate the health state. In the state update step of this filtering algorithm, the calculation of the Kalman filter gain depends on the preset prior variance of the health state and the Jacobian matrix of the observation function. When faced with complex and ever-changing time-varying operating conditions with nonlinear disturbances, this fixed state update logic lacks the ability to adaptively extract features and assign weights to the state residuals between the prior predicted state and the real-time observed state. In the long-term recursive process, the model cannot correct the prediction offset in time, which easily leads to error accumulation and state drift, ultimately resulting in the distortion of long-term lifetime prediction results. Summary of the Invention
[0005] To address the problems existing in the prior art, this invention provides a method, device, and electronic device for predicting the remaining service life of mechanical equipment under time-varying operating conditions. Its purpose is to achieve effective decoupling between degradation characteristics and operating condition characteristics, overcome the deficiency of empirical degradation fitting models lacking physical dynamic evolution constraints, and adaptively correct prediction offset based on state residuals in long-term state recursion, thereby effectively suppressing error accumulation and state drift.
[0006] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: According to a first aspect of the present invention, a method for predicting the remaining service life of mechanical equipment under time-varying operating conditions is provided, comprising: Acquire the vibration signal and operating parameters of the mechanical equipment at the current moment during operation; Based on the vibration signal and operating parameters at the current moment, the degradation information and operating condition information are separated by a pre-established decoupled health state model to obtain the observed health state and operating condition characteristics at the current moment. Using a pre-established physical degradation state space model, the posterior value of the health state at the current moment is obtained based on the posterior value of the health state at the previous moment, the operating condition characteristics at the current moment, and the observed value of the health state at the current moment. The processing steps of the physical degradation state space model include: dynamically predicting the health state at the current moment based on the posterior value of the health state at the previous moment and the operating condition characteristics at the current moment to obtain the prior value of the health state at the current moment; adaptively calculating the state update weight based on the state residual between the observed value of the health state at the current moment and the prior value of the health state at the current moment; and dynamically correcting the prior value of the health state at the current moment using the state update weight to obtain the posterior value of the health state at the current moment. The posterior value of the current health status is input into the pre-established remaining service life prediction model, and the remaining service life prediction result of the mechanical equipment is output. The physical degradation state space model is obtained by joint training of continuous-time physical degradation constraints based on the relationship between health status changes at adjacent time points, so that the evolution process of the health status posterior value finally output by the physical degradation state space model satisfies the continuous degradation law of the equipment.
[0007] In one possible implementation of the first aspect, the operating condition decoupled health state model includes a signal coding network, an operating condition coding network, and a health state coding network; The steps of obtaining the current health status observation value and the current operating condition characteristics include: The vibration signal at the current moment is input into the signal encoding network to extract the degradation features; The current operating condition parameters are input into the operating condition coding network to extract the current operating condition features. The degradation features and the current operating condition features are input into the health status coding network to generate the probability distribution parameters of the health status, and the current health status observation value is obtained through reparameterized sampling.
[0008] In one possible implementation of the first aspect, the working condition decoupled health state model is obtained by introducing variational constraints and working condition adversarial decoupling constraints for joint training, so that the health state distribution corresponding to the probability distribution parameters is close to the preset prior distribution, and the working condition interference information in the current health state observation value is suppressed.
[0009] In one possible implementation of the first aspect, the physical degradation state space model includes a dynamic prior prediction module, which includes a cyclic state update unit and a prediction compensation network. The step of dynamically predicting the health status prior value at the current moment based on the posterior value of the health status at the previous moment and the operating condition characteristics at the current moment includes: The posterior value of the health status at the previous moment and the operating condition characteristics at the current moment are input into the cyclic state update unit to obtain the basic predicted value of the health status. The predictive compensation network is used to compensate for the nonlinear degradation changes of the basic predicted value of the health status and the current working condition characteristics, and the predicted compensation amount of the health status is calculated. The influence of the health status prediction compensation amount on the basic predicted value of the health status is controlled by the compensation control coefficient, and the prior value of the health status at the current time is obtained by combining the two.
[0010] In one possible implementation of the first aspect, the physical degradation state space model further includes a posterior value update module, which includes a pre-established state gain network; The step of adaptively calculating the state update weight based on the state residual between the current health status observation value and the current health status prior value, and using the state update weight to dynamically correct the current health status prior value to obtain the current health status posterior value includes: The prior value of the current health status, the observed value of the current health status, and the operating condition features at the current time are concatenated and input into the state gain network to calculate the state update weights. After multiplying the state update weight by the corresponding elements of the state residual, the result is compensated into the prior value of the current health state to obtain the posterior value of the current health state.
[0011] In one possible implementation of the first aspect, the continuous-time physical degradation constraints used by the physical degradation state-space model during joint training are established in the following manner: The first derivative of the health status with respect to time is approximately calculated by using the finite difference between the posterior values of the health status at adjacent time points. The posterior value of the current health status and the current working condition characteristics are input into the dynamic model of continuous degradation of health status, and the predicted value of the rate of change of health status is output. The difference between the first derivative and the predicted rate of change of health status is calculated to construct the minimum continuous-time physical degradation constraint.
[0012] In one possible implementation of the first aspect, the remaining useful life prediction model includes a remaining useful life feature mapping network; The step of outputting the predicted remaining service life of the mechanical equipment includes: The posterior value of the current health status is input into the remaining lifespan feature mapping network to obtain lifespan prediction features; Based on the lifetime prediction features, the mean of the remaining lifetime prediction and the prediction uncertainty at the current moment are output respectively. Based on the predicted mean and the predicted uncertainty, a confidence interval for the remaining useful life prediction result is constructed.
[0013] In one possible implementation of the first aspect, the current operating condition parameters include at least one of rotational speed, load, and temperature; the current vibration signal is the original time-domain vibration signal, or a vibration signal after filtering, normalization, or segmentation.
[0014] According to a second aspect of the present invention, a device for predicting the remaining service life of mechanical equipment under time-varying operating conditions is provided, applied to the prediction method, the device comprising: The data acquisition module is used to acquire the vibration signal and operating parameters of the mechanical equipment at the current moment during operation. The state observation module is used to separate degradation information from working condition information based on the vibration signal and working condition parameters at the current moment, and obtain the current health status observation value and working condition characteristics at the current moment through a pre-established working condition decoupled health status model. The state tracking module is used to obtain the posterior value of the health state at the current moment based on the posterior value of the health state at the previous moment, the operating condition characteristics at the current moment, and the observed value of the health state at the current moment, using a pre-established physical degradation state space model. The processing of the physical degradation state space model includes: dynamically predicting the health state at the current moment based on the posterior value of the health state at the previous moment and the operating condition characteristics at the current moment to obtain the prior value of the health state at the current moment; adaptively calculating the state update weight based on the state residual between the observed value of the health state at the current moment and the prior value of the health state at the current moment; and dynamically correcting the prior value of the health state at the current moment using the state update weight to obtain the posterior value of the health state at the current moment. The lifespan output module is used to input the posterior value of the current health status into the pre-established remaining lifespan prediction model and output the remaining lifespan prediction result of the mechanical equipment.
[0015] The physical degradation state space model is obtained by joint training of continuous-time physical degradation constraints based on the relationship between health status changes at adjacent time points, so that the evolution process of the health status posterior value finally output by the physical degradation state space model satisfies the continuous degradation law of the equipment.
[0016] According to a third aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the aforementioned method for predicting the remaining service life of mechanical equipment under time-varying operating conditions.
[0017] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method for predicting the remaining service life of mechanical equipment under time-varying operating conditions.
[0018] According to a fifth aspect of the present invention, a computer program product is provided, which, when executed by a processor, implements the aforementioned method for predicting the remaining service life of mechanical equipment under time-varying operating conditions.
[0019] Compared with the prior art, the present invention has at least the following beneficial effects: This invention provides a method for predicting the remaining service life of mechanical equipment under time-varying operating conditions. By establishing a pre-established decoupled health state model, it separates the interference information caused by changes in operating conditions in a deep representation space, reducing the model's learning of pseudo-degradation features related to these changes. This allows for a more accurate representation of the true health state of mechanical equipment under complex operating conditions. Simultaneously, by adaptively calculating state update weights based on the state residuals between observed and prior health state values and dynamically correcting the prior health state values, it can promptly correct prediction biases caused by nonlinear disturbances, suppressing error accumulation and state drift problems in the long-term recursive process. Furthermore, the physical degradation state space model is constrained by continuous-time physical degradation based on the relationship between adjacent health state changes, ensuring that the recursive evolution of the health state conforms to the true physical degradation law continuously modulated by time-varying operating conditions, improving the model's physical consistency and long-term prediction stability. These techniques work together synergistically to ultimately improve the prediction results of the remaining service life of mechanical equipment under complex and ever-changing time-varying operating conditions.
[0020] Furthermore, by using the remaining useful life feature mapping network, the mean of the remaining useful life prediction and the prediction uncertainty at the current moment are output, and the confidence interval of the prediction result is constructed accordingly. This transforms the equipment life prediction from an absolute point estimate into a probability distribution expression, which objectively reflects the model's understanding of the current degradation trend. Attached Figure Description
[0021] To more clearly illustrate the technical solutions in the specific embodiments of the present invention, the drawings used in the description of the specific embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0022] Figure 1 This is a flowchart of a method for predicting the remaining service life of mechanical equipment under time-varying operating conditions according to the present invention.
[0023] Figure 2 The diagram shows the observed response process of mechanical equipment from operation to failure under constant and time-varying operating conditions. (a) is the observed response curve under constant operating conditions; (b) is the observed response curve under time-varying operating conditions.
[0024] Figure 3 This is a schematic diagram of the overall model framework for the remaining service life prediction method under time-varying operating conditions of the present invention.
[0025] Figure 4 This is a physical layout diagram of the rolling bearing accelerated life test platform in the embodiment.
[0026] Figure 5 The following is the rolling bearing full-life operation data under time-varying conditions in the embodiment, where (a) is the rotational speed signal time sequence curve; (b) is the vibration time domain signal; (c) is the vibration signal RMS change curve; and (d) is the rolling bearing failure type fault diagram, with each group of bearings corresponding to a sub-diagram.
[0027] Figure 6 In the embodiments, the health status estimation results obtained by the method of the present invention are shown in (a) as the health status evolution curve over time and (b) as the visualization of latent space decoupling.
[0028] Figure 7 The remaining service life prediction results of the method of the present invention in the embodiments are shown, wherein (a) is the prediction result of test bearing 1; (b) is the prediction result of test bearing 2; and (c) is the prediction result of test bearing 3. Detailed Implementation
[0029] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0030] like Figure 1As shown, the present invention provides a method for predicting the remaining service life of mechanical equipment under time-varying operating conditions, specifically addressing the scenario of predicting the remaining service life of mechanical equipment under time-varying operating conditions, and includes the following steps: S1. Obtain the vibration signal and operating parameters of the mechanical equipment at the current moment during operation.
[0031] The operating parameters include at least one of speed, load, and temperature, which can be acquired in real time by corresponding speed sensors, load sensors, temperature sensors, or the equipment control system. The vibration signal can be the original time-domain vibration signal, or a filtered, normalized, or segmented preprocessed vibration signal, acquired by a vibration acceleration sensor installed on the housing of a key component of the mechanical equipment. During the acquisition process, the vibration signal and operating parameters are synchronized according to the acquisition timestamp, and a running data sequence is constructed according to the time order, which can be represented as:
[0032] in, Indicates the total number of sampling times; For the first The vibration signal at any given time can be either the original time-domain vibration signal or the vibration signal after preprocessing. For the first The operating parameters at any given time can be either a single operating parameter or a vector composed of multiple operating parameters.
[0033] like Figure 2 As shown, the evolution patterns of monitoring signals differ significantly under different operating conditions: from Figure 2 As can be seen from (a) in the diagram, under constant operating conditions, the equipment's operating conditions remain fixed, and the health state monotonically evolves from healthy to faulty. The monitoring signal is determined only by the equipment's health state and the observed noise, and the corresponding expression is: The signal amplitude monotonically increases with the degradation process, without additional periodic fluctuations, and the signal change trend directly corresponds to the degradation trend; from Figure 2 As can be seen from (b) in the figure, under time-varying operating conditions, the operating parameters change dynamically with time, and the monitoring signal is simultaneously modulated by both health status degradation and time-varying operating conditions. The corresponding expression is: The signal exhibits periodic fluctuations synchronized with changes in operating conditions. The degradation information in the original signal is deeply coupled with the operating condition information, making it impossible to directly determine the true degree of equipment degradation through the signal amplitude.
[0034] S2. Based on the vibration signal and operating parameters at the current moment, the degradation information and operating condition information are separated by a pre-established decoupled health state model to obtain the current health state observation value and the current operating condition characteristics.
[0035] Specifically, the operating condition decoupled health state model belongs to the encoder module of the overall model, including the signal encoding network, the operating condition encoding network, and the health state encoding network.
[0036] The current vibration signal is input into a signal encoding network, and the degradation features are extracted, which can be represented as:
[0037] in, The signal coding network, for example, see Table 1 below, uses a lightweight structure of 3 one-dimensional convolutional layers combined with global average pooling layers and fully connected layers to automatically extract degradation-related features from the original vibration signal; Indicates the first The degradation characteristics of time, Indicates the first Vibration signal at any given moment.
[0038] Simultaneously, the current operating condition parameters are input into the operating condition coding network to extract the current operating condition features, which can be represented as:
[0039] in, The operating condition coding network, for example, see Table 1 below, adopts a 2-layer fully connected layer structure to map operating condition parameters to a high-dimensional operating condition feature space; Indicates the first Operating characteristics at any given time, Indicates the first Operating parameters at any given time.
[0040] Furthermore, the degradation features and the current operating condition features are input together into the health status coding network to generate probability distribution parameters of the health status, and to model the probability distribution of the health status:
[0041]
[0042] in, For the health status encoding network, the concatenated vector of degradation features and working condition features is used as input. For example, see Table 1 below. A two-layer fully connected layer structure is used to output the probability distribution parameters. Indicates the first Real-time device health status Indicates a Gaussian distribution. and Let represent the mean and standard deviation of the health status distribution, respectively.
[0043] To obtain health state observations that can be used for subsequent state estimation, this implementation method samples the health state distribution through reparameterization to obtain the health state observations at the current time:
[0044] in, Indicates the first Health status observations at any given time Let be a random variable that follows a standard normal distribution. This indicates an element-wise multiplication operation.
[0045] The decoupled health state model under this operating condition is obtained through joint training by introducing variational constraints and adversarial decoupling constraints. During training, the health state distribution is made to approximate a pre-defined standard normal prior distribution through variational constraints, corresponding to the following variational regularization term:
[0046] This constraint can improve the continuity and stability of the health state space and avoid discretization jumps in the state space.
[0047] Meanwhile, an adversarial decoupling constraint for operating conditions is constructed through a gradient inversion mechanism. An operating condition classifier based on a gradient inversion layer is introduced into the health state space to form an adversarial training mechanism with the health state encoding network: the operating condition classifier attempts to identify the operating condition category from the health state, while the health state encoding network actively suppresses the operating condition information in the health state to mislead the classifier. Ultimately, the health state observation values retain degradation information while weakening the interference of operating conditions, thus obtaining a health state representation with operating condition invariance. This reduces the interference of operating condition fluctuations on the degradation representation from the feature level.
[0048] The above encoding and decoupling processes correspond to Figure 3 The operating condition decoupling health encoder module in the overall framework. From... Figure 3 It can be seen from the vibration signal Operating parameters Features are extracted by two independent coding branches and then fused to generate distribution parameters of health status observations. At the same time, weak variational regularization constraints are introduced to constrain the health status distribution. The overall structure realizes the coding separation of degradation information and operating condition information.
[0049] S3. Using a pre-established physical degradation state space model, based on the posterior value of the health state at the previous moment, the operating condition characteristics at the current moment, and the observed value of the health state at the current moment, obtain the posterior value of the health state at the current moment.
[0050] Specifically, the physical degradation state space model includes a dynamic prior prediction module and a posterior update module. In this embodiment, the physical degradation state space model is the Physics-informed Neural State Space Model (PINSSM), which belongs to the state space filtering module of the overall model. The dynamic prior prediction module includes a cyclic state update unit and a prediction compensation network, and the posterior update module includes a state gain network.
[0051] During the processing of the dynamic prior prediction module, dynamic prediction is performed based on the posterior value of the health status at the previous moment and the operating condition characteristics at the current moment to obtain the prior value of the health status at the current moment, as follows: Let the first The posterior value of health status at any given time Then the first The prior value of health status at any given time can be represented as:
[0052] in, This represents the state transition function, used to describe the dynamic evolution of the health state under the influence of operating condition modulation.
[0053] First, the posterior value of the health status at the previous time step and the operating condition characteristics at the current time step are input into the cyclic state update unit to obtain the basic predicted value of the health status. In this embodiment, the cyclic state update unit is implemented using a gated recurrent unit (GRU), and the corresponding expression is:
[0054] in, Indicates the first Baseline predicted health status at any given time. Indicates the first Posterior value of health status at any given time.
[0055] The predictive compensation network is used to compensate for the nonlinear degradation changes in the baseline predicted health status and the current operating condition characteristics, and the predicted health status compensation amount is calculated:
[0056] in, The prediction compensation network, also known as the residual mapping network, takes the concatenated vector of the basic predicted health status and the operating condition characteristics as input. For example, see Table 1 below. It adopts a 2-layer fully connected layer structure to compensate for nonlinear degradation changes under complex time-varying operating conditions. This represents a vector concatenation operation. This indicates the amount of compensation predicted for health status.
[0057] Finally, the impact of the health status prediction compensation on the basic predicted value of health status is controlled by the compensation control coefficient, and the prior value of health status at the current time is obtained by combining the results.
[0058] in, Indicates the first Prior values of health status at any given time; This represents the compensation control coefficient, used to control the impact of the health status prediction compensation on the health status prediction results, and adapting to the changes in degradation rate under different operating conditions.
[0059] During the processing of the posterior value update module, the state update weight is adaptively calculated based on the state residual between the current health status observation value and the current health status prior value. The state update weight is then used to dynamically correct the current health status prior value to obtain the current health status posterior value, as detailed below: First, calculate the state residual between the current observed health status and the prior health status, which is used to characterize the deviation between real-time observation information and historical prediction results:
[0060] in, Indicates the first The state residual at any given moment.
[0061] Then, the prior health status value, the observed health status value, and the operating condition characteristics at the current time are concatenated and input into a pre-established state gain network to calculate the state update weights:
[0062] in, The state gain network takes a concatenated vector of health state prior values, health state observation values, and operating condition features as input. For example, see Table 1 below. It is implemented using a 2-layer fully connected layer structure. Indicates the first The state updates the weights at each moment.
[0063] Finally, the state update weights and state residuals are multiplied element-wise and then added to the prior value of the current healthy state to obtain the posterior value of the current healthy state.
[0064] in, Indicates the first Posterior value of health status at any given time.
[0065] This adaptive correction mechanism can dynamically adjust the fusion weights of observation information according to the current operating conditions and the degree of state deviation. Compared with the fixed gain update logic that relies on preset parameters in traditional filtering algorithms, it can correct the prediction offset in a timely manner when the operating conditions fluctuate significantly, effectively suppressing the accumulation of errors and state drift in the long-term recursive process.
[0066] from Figure 3 As can be seen, the PINSSM module operates in a time-series recursive manner. At each time step, it receives the posterior state of the previous time step, the current operating condition, and the current observation value, and completes the complete state update process of prediction-correction. The output posterior health status integrates historical degradation trends and real-time observation information.
[0067] The physical degradation state-space model is obtained through joint training of continuous-time physical degradation constraints based on the relationship between health state changes at adjacent time points. This ensures that the evolution of the posterior health state value output by the physical degradation state-space model conforms to the continuous degradation law of the equipment. The continuous-time physical degradation constraints are established in the following way: During training, the first derivative of the health state with respect to time is approximately calculated using the finite difference between the posterior values of the health state at adjacent time points:
[0068] in, This represents the time interval between two adjacent sampling times; Input the posterior value of the current health status and the current operating condition characteristics into the dynamic model of continuous health status degradation, and output the predicted value of the rate of change of health status:
[0069] in, The dynamic model of continuous degradation of health status is represented by a concatenated vector of health status posterior values and working condition characteristics as input. For example, see Table 1 below. It is implemented using a 3-layer fully connected layer structure. The difference between the first derivative and the predicted rate of change in health status is calculated, and a continuous-time physical degradation constraint is constructed to minimize this difference. :
[0070] By minimizing this constraint term during joint training, abnormal fluctuations in the health state during the long-term recursive process can be suppressed, making the evolution trend of the health state conform to the real physical law of continuous cumulative degradation of mechanical equipment performance.
[0071] It should be noted that the continuous-time physical degradation constraint established in this embodiment does not depend on the specific structural parameters of the equipment or complex analytical physical models, but rather constructs a unified degradation constraint based on the continuous evolution law of the health state, thus having good applicability.
[0072] S4. Input the posterior value of the current health status into the pre-established remaining service life prediction model, and output the prediction result of the remaining service life of the mechanical equipment.
[0073] Specifically, the remaining useful life prediction model includes a remaining useful life feature mapping network and a Gaussian output layer, corresponding to the Gaussian RUL prediction head in the overall model decoder module. The posterior value of the current health status is input into the remaining useful life feature mapping network to obtain the lifespan prediction features:
[0074] in, The feature mapping network representing the remaining lifetime is implemented using a two-layer fully connected layer structure. This indicates lifespan prediction characteristics.
[0075] Based on the lifetime prediction characteristics, the mean of the remaining lifetime prediction and the prediction uncertainty at the current moment are output respectively:
[0076]
[0077] in, Indicates the first The predicted remaining useful life at any given time. This indicates the uncertainty corresponding to the remaining useful life prediction results. , These are the weight parameters and bias parameters corresponding to the predicted mean, respectively. , These are the weight parameters and bias parameters corresponding to the prediction uncertainty, respectively.
[0078] The predicted final remaining useful life (RUL) can be expressed as a probability form that follows a Gaussian distribution:
[0079] Based on the predicted mean and the predicted uncertainty, a confidence interval for the remaining useful life prediction results can be constructed. In this embodiment, a 95% confidence interval is constructed to quantify the reliability of the prediction results and provide a more comprehensive reference for equipment condition assessment and maintenance decisions.
[0080] The aforementioned decoupled health state model, physical degradation state space model, and remaining useful life prediction model are trained using an end-to-end joint optimization approach. During training, the accompanying decoder module participates in the optimization synchronously to support the calculation of various constraints. The overall joint optimization objective function can be expressed as:
[0081] in, The health status indicates the optimization goal. This represents a continuous-time physical degradation constraint. Indicates the target for remaining useful life prediction. and These represent the weight coefficients for the corresponding objectives.
[0082] The optimization objective for health status representation aims to reduce the impact of operating condition changes on the health status representation. Specifically, it consists of feature reconstruction constraints, variational regularization constraints, operating condition feature preservation constraints, and operating condition adversarial decoupling constraints.
[0083] For example, see Table 1 below, where, The feature reconstruction constraint is represented by the signal decoder, which reconstructs the original degradation features based on the health status and operating condition features. The signal decoder takes the concatenated vector of the health status posterior value and the operating condition features as input and outputs the reconstructed degradation features using a 2-layer fully connected layer structure. This constraint is used to ensure that the health status retains complete degradation information. This represents a variational regularization constraint, used to constrain the distribution of health states; This indicates that the operating condition features are maintained as a constraint. The operating condition decoder predicts the operating condition category based on the operating condition features. The operating condition decoder takes the operating condition features as input and adopts a structure of two fully connected layers combined with a Softmax activation layer to output the predicted operating condition category. This constraint is used to ensure that the operating condition coding network can effectively extract operating condition information. This indicates that the working condition is decoupled from the working condition. The gradient inversion layer classifier predicts the working condition category based on the health state. The gradient inversion layer classifier takes the posterior value of the health state as input and adopts a structure of two fully connected layers combined with a Softmax activation layer to output the working condition category prediction result. This constraint is used to suppress the working condition information in the healthy state. , , , These are the weighting coefficients for the corresponding loss terms.
[0084] The remaining useful life prediction objective employs Gaussian negative log-likelihood loss to simultaneously optimize the fitting accuracy of the prediction mean and uncertainty:
[0085] in, Indicates the first The label shows the actual remaining service life at any given time. This indicates the number of training samples.
[0086] In this embodiment, training data organized according to time series is input into each component module during the training process. The training data comes from multiple sets of full-lifetime operation data. Each set of full-lifetime operation data is sampled by sliding within a fixed time window. Each sample contains a vibration signal sequence { }、Working condition parameter sequence{ } and the corresponding real URI tags { },in Defined as the normalized value obtained by dividing the remaining time from the current moment to the failure moment by the total lifetime, ensuring that the label value range is between [0,1]. The training data is divided into a training set and a validation set, randomly partitioned in an 8:2 ratio to ensure that samples from different degradation stages and operating condition change patterns are covered in both sets. The weight coefficients of each loss term are set as follows: , = , , , , The Adam optimizer synchronously updates all network parameters, with an initial learning rate of 1e-3, a batch size of 32, and a maximum training epoch of 100. An early stopping strategy is employed to prevent overfitting. To improve the model's adaptability to samples of different observation lengths, a random subsequence sampling strategy is used during training: for samples with short-cycle rotational speed changes, the sampling window length is set to [32, 64]; for samples with longer-cycle rotational speed changes, the sampling window length is set to [160, 180]. Training stops when the joint optimization objective function converges or reaches the preset number of training epochs (for example, training terminates when the validation set loss does not decrease for 10 consecutive epochs), resulting in the final model for predicting the remaining service life of the mechanical equipment. After training, all network parameters, including those of the encoder, state-space filter module, decoder, and prediction head, are fixed and saved for fixed forward computation during the online inference phase. During continuous equipment operation, the process from S1 to S4 is repeated according to the sampling time, enabling continuous tracking of health status and online dynamic prediction of remaining service life.
[0087] The detailed network structure is summarized in Table 1.
[0088] Table 2
[0089] Compared to existing technologies, this solution achieves a comprehensive improvement in prediction performance through three core improvements. First, it employs a deep dual-branch coding network combined with adversarial decoupling constraints to construct a decoupled health state model. This model can separate pseudo-degradation information caused by operating condition fluctuations in a high-dimensional feature space, ensuring that the health state representation only reflects the true degree of equipment degradation. This fundamentally reduces the interference of time-varying operating conditions on state estimation and improves the accuracy of degradation state representation under complex operating conditions. Second, it constructs a physical degradation state space model that integrates cyclic temporal units and nonlinear compensation, and introduces continuous-time physical degradation constraints to participate in end-to-end training. This anchors the discrete state recursion process to the physical laws of continuous degradation, accurately describing the dynamic evolution of degradation under operating condition modulation, enhancing the physical consistency of degradation modeling, and improving the stability of the long-term prediction process. Third, a fully data-driven adaptive state gain network is used to achieve state correction. The observation fusion weights are dynamically adjusted based on real-time state residuals and operating condition information, replacing the fixed gain update logic of traditional filtering algorithms that rely on preset parameters. This effectively suppresses the accumulation of errors and state drift in long-term recursion, and further improves the accuracy of life prediction results.
[0090] To verify the practical effectiveness of the above method, a rolling bearing accelerated life test platform was used to conduct verification experiments. For example... Figure 4 As shown, the test platform mainly consists of an AC motor, a support shaft, support bearings, test bearings, a hydraulic loading device, a photoelectric tachometer, an accelerometer, a data acquisition system, and a host computer. The test object is an ER-16K deep groove ball bearing. During the test, radial and axial loads are applied through the hydraulic loading system, the rotational speed is measured in real time through the photoelectric tachometer, and the vibration response signal is collected through the accelerometer installed on the bearing housing.
[0091] The test setup involved a continuous periodic variation of the rotational speed throughout its entire lifespan, adjusted every 0.5 hours with a variation amplitude of 10 Hz. The first 0.1 seconds of each acquired signal were used as model input, and the corresponding rotational speed was also input as the operating condition parameter into the model. Three sets of test bearings were used, with specific test parameters shown in Table 2 below: Bearing 1 was subjected to a radial force of 4.0 kN and an axial force of 0 kN; Bearing 2 was subjected to a radial force of 3.5 kN and an axial force of 0.5 kN; Bearing 3 was subjected to a radial force of 3.5 kN and an axial force of 0.5 kN. The sampling frequency for all three sets of tests was 20480 Hz, with vibration signals acquired every 10 minutes for a duration of 3 seconds.
[0092] Table 2
[0093] Rolling bearing full life operating data under time-varying conditions, such as Figure 5 As shown: from Figure 5 As can be seen from the speed signal curves (a1), (a2), and (a3), the three bearings correspond to different lifespans and speed fluctuation periods: (a1) corresponds to test bearing 1, with the speed fluctuating periodically in the range of 10~50Hz, and approximately 240 sampling points throughout the entire lifespan; (a2) corresponds to test bearing 2, with a higher speed change frequency, approximately 600 sampling points throughout the entire lifespan, and a longer degradation process; (a3) corresponds to test bearing 3, with a speed fluctuation period similar to that of test bearing 1, and approximately 300 sampling points throughout the entire lifespan.
[0094] from Figure 5 The time-domain waveforms of the original vibration signals (b1), (b2), and (b3) show that the vibration signal amplitudes of the three bearings exhibit synchronous periodic changes with the rotational speed fluctuations. At the same time, the overall amplitude gradually increases with the deepening of degradation, which intuitively reflects the coupling characteristics of degradation information and working condition information. Among them, the long-cycle life sample corresponding to (b2) has a low vibration amplitude and stable fluctuations in the early stage, and the amplitude increases significantly in the later failure stage.
[0095] from Figure 5 The root mean square (RMS) curves of the vibration signals (c1), (c2), and (c3) show that all three RMS curves exhibit obvious periodic sawtooth fluctuations with changes in rotational speed, which cannot directly reflect the true degradation trend of the bearing and verify the limitations of directly predicting life based on statistical characteristics. As degradation intensifies, the overall baseline of RMS gradually rises, with the bearing corresponding to (c2) showing a sharp increase in RMS at the end of its life, corresponding to the rapid failure stage of the bearing.
[0096] from Figure 5 The physical images of failure types (d1), (d2), and (d3) show that the test samples cover different failure modes: (d1) and (d2) represent inner ring failures, with obvious wear and spalling areas appearing on the inner ring raceway; (d3) represents cage failure, with the cage showing fracture and deformation. This fully verifies the generalization ability of the method under different failure modes. The model's generalization ability is verified using leave-one-out cross-validation, that is, two sets of bearing data are selected as the training set each time, and the remaining set of bearing data is used as the test set, and all tests are completed alternately.
[0097] Health status estimation results as follows Figure 6 As shown: from Figure 6The health status evolution curves (a1), (a2), and (a3) in the top row show that the orange curve represents the first principal component of the posterior value of the health status, and the light blue background represents the normalized fluctuation amplitude of the operating conditions. Although the operating conditions exhibit periodic and significant fluctuations, the overall health status curve maintains a monotonically increasing degradation trend and does not fluctuate synchronously with the operating conditions, verifying the effectiveness of the operating condition decoupling. The health status remains stable in the early stage of degradation and continues to rise after entering the accelerated degradation stage, which is consistent with the physical laws of bearing degradation. Among them, the long-life sample corresponding to (a2) shows that the health status remains low and stable for a long time in the early stage and rises rapidly in the final stage, which is consistent with the actual degradation process.
[0098] from Figure 6 The visualization results of latent space decoupling in the lower row (b1)(b2)(b3) show that the posterior health state exhibits a clear and continuous degradation trajectory in the view marked by degradation, while no clustering phenomenon dominated by operating conditions appears in the view marked by operating conditions. Conversely, the latent variables of operating conditions form clearly separable clusters in the view marked by operating conditions, while no obvious degradation evolution trend is observed in the view based on degradation coloring. These results indicate that the decoupling framework proposed in this invention successfully encodes degradation information and operating condition information into the posterior health state and latent variables of operating conditions, respectively, achieving effective separation of health state and operating conditions.
[0099] Remaining useful life prediction results are as follows Figure 7 As shown, the three sub-graphs correspond to three groups of test bearings: from Figure 7 As can be seen from (a) in the figure, the predicted RUL curve of the test bearing 1 has a high degree of fit with the real RUL, the 95% confidence interval covers the real RUL curve in its entirety, the prediction error is small, and the trend following is good.
[0100] from Figure 7 As can be seen from (b) in the figure, the test bearing 2 belongs to the long-term degradation sample. The prediction curve did not show obvious drift in the long-term recursion process and can accurately track the linear downward trend of RUL, which verifies the long-term prediction stability of the method.
[0101] from Figure 7 As can be seen from (c), the predicted curve of the test bearing 3 in the middle to the end of the degradation process is in high agreement with the actual value. The confidence interval width changes reasonably with the degradation process, which can reflect the prediction uncertainty at different stages.
[0102] The quantitative evaluation indicators in Table 3 further demonstrate the accuracy of the RUL prediction using this method.
[0103] Table 3
[0104] The quantitative evaluation indicators include three items: RMSE, DWT, and Score. The evaluation results for the three groups of tested bearings are as follows: Bearing 1 has an RMSE of 0.042, a DWT of 3.244, and a Score of 0.664; Bearing 2 has an RMSE of 0.079, a DWT of 4.963, and a Score of 2.295; Bearing 3 has an RMSE of 0.042, a DWT of 3.754, and a Score of 0.726. The above qualitative and quantitative results jointly verify that this method can effectively improve the prediction accuracy and stability of the remaining service life of mechanical equipment under complex time-varying operating conditions.
[0105] In another embodiment of the present invention, a device for predicting the remaining service life of mechanical equipment under time-varying operating conditions is provided, applied to the prediction method described above. The device includes: The data acquisition module is used to acquire the vibration signal and operating parameters of the mechanical equipment at the current moment during operation. The state observation module is used to separate degradation information from working condition information based on the vibration signal and working condition parameters at the current moment, and obtain the current health status observation value and working condition characteristics at the current moment through a pre-established working condition decoupled health status model. The state tracking module is used to obtain the posterior value of the health state at the current moment based on the posterior value of the health state at the previous moment, the operating condition characteristics at the current moment, and the observed value of the health state at the current moment, using a pre-established physical degradation state space model. The processing of the physical degradation state space model includes: dynamically predicting the health state at the current moment based on the posterior value of the health state at the previous moment and the operating condition characteristics at the current moment to obtain the prior value of the health state at the current moment; adaptively calculating the state update weight based on the state residual between the observed value of the health state at the current moment and the prior value of the health state at the current moment; and dynamically correcting the prior value of the health state at the current moment using the state update weight to obtain the posterior value of the health state at the current moment. The lifespan output module is used to input the posterior value of the current health status into the pre-established remaining lifespan prediction model and output the remaining lifespan prediction result of the mechanical equipment. The physical degradation state space model is obtained by joint training of continuous-time physical degradation constraints based on the relationship between health status changes at adjacent time points, so that the evolution process of the health status posterior value finally output by the physical degradation state space model satisfies the continuous degradation law of the equipment.
[0106] All relevant content regarding the steps involved in the aforementioned embodiment of a method for predicting the remaining service life of mechanical equipment under time-varying operating conditions can be referenced from the functional description of the corresponding functional module of the device for predicting the remaining service life of mechanical equipment under time-varying operating conditions in this embodiment of the invention, and will not be repeated here. The module division in the embodiments of the invention is illustrative and is merely a logical functional division. In actual implementation, there may be other division methods. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a processor, exist separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module.
[0107] In another embodiment of the present invention, an electronic device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions from the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used in the operation of a method for predicting the remaining service life of mechanical equipment under time-varying operating conditions.
[0108] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in an electronic device used to store programs and data. It is understood that the computer-readable storage medium here can include both built-in storage media in the electronic device and extended storage media supported by the electronic device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be Random Access Memory (RAM) or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the method for predicting the remaining service life of mechanical equipment under time-varying operating conditions in the above embodiments.
[0109] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.
[0110] This invention is described 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 program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0111] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0112] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0113] This invention also provides a computer program product for executing any of the above-described methods for predicting the remaining service life of mechanical equipment under time-varying operating conditions. Since the computer program product provided by this invention belongs to the same inventive concept as the above-described method for predicting the remaining service life of mechanical equipment under time-varying operating conditions, it possesses all the advantages of the above-described method. Therefore, the beneficial effects of the computer program product provided by this invention will not be elaborated upon here.
[0114] In this invention, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0115] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the scope of the technology disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention.
Claims
1. A method for predicting the remaining service life of mechanical equipment under time-varying operating conditions, characterized in that, include: Acquire the vibration signal and operating parameters of the mechanical equipment at the current moment during operation; Based on the vibration signal and operating parameters at the current moment, the degradation information and operating condition information are separated by a pre-established decoupled health state model to obtain the observed health state and operating condition characteristics at the current moment. Using a pre-established physical degradation state space model, the posterior value of the health state at the current moment is obtained based on the posterior value of the health state at the previous moment, the operating condition characteristics at the current moment, and the observed value of the health state at the current moment. The processing steps of the physical degradation state space model include: dynamically predicting the health state at the current moment based on the posterior value of the health state at the previous moment and the operating condition characteristics at the current moment to obtain the prior value of the health state at the current moment; adaptively calculating the state update weight based on the state residual between the observed value of the health state at the current moment and the prior value of the health state at the current moment; and dynamically correcting the prior value of the health state at the current moment using the state update weight to obtain the posterior value of the health state at the current moment. The posterior value of the current health status is input into the pre-established remaining service life prediction model, and the remaining service life prediction result of the mechanical equipment is output. The physical degradation state space model is obtained by joint training of continuous-time physical degradation constraints based on the relationship between health status changes at adjacent time points, so that the evolution process of the health status posterior value finally output by the physical degradation state space model satisfies the continuous degradation law of the equipment.
2. The method for predicting the remaining service life of mechanical equipment under time-varying operating conditions according to claim 1, characterized in that, The decoupled operating condition health state model includes a signal coding network, an operating condition coding network, and a health state coding network. The steps of obtaining the current health status observation value and the current operating condition characteristics include: The vibration signal at the current moment is input into the signal encoding network to extract the degradation features; The current operating condition parameters are input into the operating condition coding network to extract the current operating condition features. The degradation features and the current operating condition features are input into the health status coding network to generate the probability distribution parameters of the health status, and the current health status observation value is obtained through reparameterized sampling.
3. The method for predicting the remaining service life of mechanical equipment under time-varying operating conditions according to claim 2, characterized in that, The decoupled health state model is obtained by introducing variational constraints and decoupled working conditions constraints for joint training, so that the health state distribution corresponding to the probability distribution parameters is close to the preset prior distribution, and the working condition interference information in the current health state observation value is suppressed.
4. The method for predicting the remaining service life of mechanical equipment under time-varying operating conditions according to claim 1, characterized in that, The physical degradation state space model includes a dynamic prior prediction module, which includes a cyclic state update unit and a prediction compensation network. The step of dynamically predicting the health status prior value at the current moment based on the posterior value of the health status at the previous moment and the operating condition characteristics at the current moment includes: The posterior value of the health status at the previous moment and the operating condition characteristics at the current moment are input into the cyclic state update unit to obtain the basic predicted value of the health status. The predictive compensation network is used to compensate for the nonlinear degradation changes of the basic predicted value of the health status and the current working condition characteristics, and the predicted compensation amount of the health status is calculated. The influence of the health status prediction compensation amount on the basic predicted value of the health status is controlled by the compensation control coefficient, and the prior value of the health status at the current time is obtained by combining the two.
5. The method for predicting the remaining service life of mechanical equipment under time-varying operating conditions according to claim 1, characterized in that, The physical degradation state space model also includes a posterior value update module, which includes a pre-established state gain network. The step of adaptively calculating the state update weight based on the state residual between the current health status observation value and the current health status prior value, and using the state update weight to dynamically correct the current health status prior value to obtain the current health status posterior value includes: The prior value of the current health status, the observed value of the current health status, and the operating condition features at the current time are concatenated and input into the state gain network to calculate the state update weights. After multiplying the state update weight by the corresponding elements of the state residual, the result is compensated into the prior value of the current health state to obtain the posterior value of the current health state.
6. The method for predicting the remaining service life of mechanical equipment under time-varying operating conditions according to claim 1, characterized in that, The continuous-time physical degradation constraints used in the joint training process of the physical degradation state-space model are established in the following manner: The first derivative of the health status with respect to time is approximately calculated by using the finite difference between the posterior values of the health status at adjacent time points. The posterior value of the current health status and the current working condition characteristics are input into the dynamic model of continuous degradation of health status, and the predicted value of the rate of change of health status is output. The difference between the first derivative and the predicted rate of change of health status is calculated to construct the minimum continuous-time physical degradation constraint.
7. The method for predicting the remaining service life of mechanical equipment under time-varying operating conditions according to claim 1, characterized in that, The remaining useful life prediction model includes a remaining useful life feature mapping network; The step of outputting the remaining service life prediction result of the mechanical equipment includes: The posterior value of the current health status is input into the remaining lifespan feature mapping network to obtain lifespan prediction features; Based on the lifetime prediction features, the mean of the remaining lifetime prediction and the prediction uncertainty at the current moment are output respectively. Based on the predicted mean and the predicted uncertainty, a confidence interval for the remaining useful life prediction result is constructed.
8. The method for predicting the remaining service life of mechanical equipment under time-varying operating conditions according to any one of claims 1 to 7, characterized in that, The operating parameters at the current moment include at least one of rotational speed, load, and temperature; the vibration signal at the current moment is the original time-domain vibration signal, or a vibration signal after filtering, normalization, or segmentation.
9. A device for predicting the remaining service life of mechanical equipment under time-varying operating conditions, characterized in that, The apparatus used in the prediction method according to any one of claims 1 to 8 comprises: The data acquisition module is used to acquire the vibration signal and operating parameters of the mechanical equipment at the current moment during operation. The state observation module is used to separate degradation information from working condition information based on the vibration signal and working condition parameters at the current moment, and obtain the current health status observation value and working condition characteristics at the current moment through a pre-established working condition decoupled health status model. The state tracking module is used to obtain the posterior value of the health state at the current moment based on the posterior value of the health state at the previous moment, the operating condition characteristics at the current moment, and the observed value of the health state at the current moment, using a pre-established physical degradation state space model. The processing of the physical degradation state space model includes: dynamically predicting the health state at the current moment based on the posterior value of the health state at the previous moment and the operating condition characteristics at the current moment to obtain the prior value of the health state at the current moment; adaptively calculating the state update weight based on the state residual between the observed value of the health state at the current moment and the prior value of the health state at the current moment; and dynamically correcting the prior value of the health state at the current moment using the state update weight to obtain the posterior value of the health state at the current moment. The lifespan output module is used to input the posterior value of the current health status into the pre-established remaining lifespan prediction model and output the remaining lifespan prediction result of the mechanical equipment. The physical degradation state space model is obtained by joint training of continuous-time physical degradation constraints based on the relationship between health status changes at adjacent time points, so that the evolution process of the health status posterior value finally output by the physical degradation state space model satisfies the continuous degradation law of the equipment.
10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements a method for predicting the remaining service life of mechanical equipment under time-varying operating conditions as described in any one of claims 1 to 8.
Citation Information
Patent Citations
Integrated method and device for time-varying working condition anomaly detection and residual service life prediction of rotating machinery
CN121365338A
