Integrated method and device for time-varying working condition anomaly detection and residual service life prediction of rotating machinery

Abstract

The invention discloses an integrated method and device for time-varying working condition anomaly detection and residual service life prediction of a rotating machine, and the method comprises the steps: firstly obtaining a root-mean-square value and a rotating speed of a vibration signal of the rotating machine in real time, initializing a two-stage mixed state space model which comprises a state equation and an observation equation, the state transition coefficient determines that the model is in a healthy or degeneration stage, and the observation equation adopts a neural network as a nonlinear observation function. In a health stage, a state transition coefficient is set to be 1, current data is input to calculate neural network parameters, an anomaly detection index is constructed after dimension reduction, and real-time detection is performed in combination with a statistical confidence criterion. If no abnormity is detected, returning to continue processing; and if abnormity is detected, entering a degradation stage, inputting current data to calculate the health state of the rotating machine, selecting an optimal degradation fitting model according to a model selection criterion, and updating a state transition coefficient. And finally, based on the updated model, predicting residual service life probability density distribution by using a numerical simulation technology. According to the method, the health state of the rotating machine serves as a direct solution variable, and anomaly detection, health state tracking and RUL accurate prediction are achieved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of mechanical equipment fault detection, and particularly relates to a rotating machinery time-varying working condition anomaly detection and residual service life prediction integrated method and equipment. BACKGROUND

[0002] In modern mechanical equipment, rolling bearings are one of the core components to ensure the normal operation of the equipment, and the health state thereof directly determines the safety and stability and operation reliability of the whole machine. In actual application, rolling bearings often need to serve in harsh working conditions such as high temperature, alternating load, time-varying speed, etc. for a long time. Even if the daily maintenance is in place, the risk of failure is difficult to completely avoid. Once the bearing fails, it may cause the equipment to stop running, or even cause production interruption or safety accidents. Therefore, it is a key requirement in the field of industrial equipment operation and maintenance to carry out effective prediction and health management (PHM) of rolling bearings, to identify the failure risk and predict the residual life in advance. In the PHM technical system, anomaly detection and residual useful life (RUL) prediction are two core links, which have attracted widespread attention in both academic research and industrial application in recent years. Accurate anomaly detection can timely find early signs of bearing degradation, and reliable RUL prediction can quantify the remaining service time of the bearing, and the combination of the two can provide a complete basis for the health state evaluation of the equipment.

[0003] The current mainstream RUL prediction method based on vibration signals still has obvious industrial applicability. On the one hand, most of such methods rely on a large amount of historical fault data to carry out model training or parameter calibration, but in actual industrial scenarios, in order to avoid bearing failure leading to production line shutdown or safety accidents, enterprises usually replace the bearing before obvious degradation occurs, and it is difficult to accumulate enough number of fault data covering the complete degradation process. On the other hand, existing researches are mostly based on the ideal assumption that the equipment is in a constant working condition, and it is believed that under this condition, the vibration signal of the mechanical system can clearly reflect the bearing degradation trend, but in the real industrial environment, the working condition of the rolling bearing is often time-varying, such as the speed fluctuation with process switching during machine tool machining, the load change with material quantity during conveying equipment, and the speed change due to air volume adjustment during fan operation. Under this condition, the vibration signal not only contains the health state information of the bearing, but also is covered by the interference signal caused by the working condition change, which leads to the degradation characteristics being submerged and affects the accuracy of RUL prediction.

[0004] For the bearing health monitoring problem under time-varying working conditions, a related method is disclosed in Chinese Patent Publication No. CN115219198A. The scheme proposes a hybrid state space model by introducing working condition information into the model design, which to some extent solves the influence of working condition interference on health monitoring, and provides a feasible idea for bearing state monitoring under time-varying working conditions. However, this technical solution still has obvious limitations: it does not take the bearing health state as a directly solvable variable in the model, and can only achieve anomaly detection through the intermediate output results of the neural network; once an anomaly is detected, if further tracking of the dynamic changes of the bearing health state or prediction of RUL based on the current state is required, this model cannot meet the demand. SUMMARY

[0005] To solve the problems in the prior art, the present application provides a rotating machinery time-varying working condition anomaly detection and remaining useful life prediction integrated method and device, which takes the rotating machinery health state as a directly solvable variable in the model, not only can realize anomaly detection, but also can further track the dynamic changes of the rotating machinery health state after detecting the anomaly, and accurately predict RUL based on the current state, thereby providing effective anomaly detection and remaining useful life integrated prediction for rotating machinery under time-varying working conditions.

[0006] To solve the above technical problems, the present application is implemented by the following technical solutions: According to the first aspect of the present application, a rotating machinery time-varying working condition anomaly detection and remaining useful life prediction integrated method is provided, comprising the following steps: S1, acquiring the root mean square value of the vibration signal and the rotating speed in the running process of the rotating machinery in real time; S2, initializing a two-stage hybrid state space model, the two-stage hybrid state space model being used to represent the functional relationship between the rotating speed, the health state and the root mean square value of the vibration signal in the whole life cycle of the rotating machinery, comprising a state equation and an observation equation, wherein the state transition coefficient in the state equation determines whether the two-stage hybrid state space model is in a health stage or a degradation stage, and the observation equation adopts a neural network as a nonlinear observation function; S3, in the health stage, setting the state transition coefficient to 1, and inputting the root mean square value of the vibration signal and the rotating speed at the current time into the two-stage hybrid state space model, and calculating the neural network parameters at the current time by using a nonlinear filtering algorithm; S4, performing dimensionality reduction processing on the neural network parameters, constructing a real-time anomaly detection index based on the dimensionality reduction result, and combining a statistical confidence criterion to perform real-time anomaly detection on the rotating machinery; S5, if no anomaly is detected, returning to S3 to continue the health stage processing; if an anomaly is detected, entering S6; S6. During 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 the health state of the rotating machinery at the current moment is calculated using a nonlinear filtering algorithm. S7. Based on the current time and the health status before the current time, select the best degradation fitting model from a number of preset degradation fitting models based on the model selection criteria, and update the state transition coefficients based on the best degradation fitting model. S8. Based on the two-stage hybrid state-space model after updating the state transition coefficients, numerical simulation technology is used to predict the probability density distribution of the remaining service life of the rotating machinery.

[0007] In one possible implementation of the first aspect, the specific form of the two-stage hybrid state-space model is as follows: Equations of state:

[0008] Observation equation:

[0009] in,

[0010]

[0011] In the formula: For rotating machinery Constant health status; For rotating machinery Constant health status; For rotating machinery State transition coefficient at time t; State noise; The best-fit model for degradation; and These are the parameters in the best-fit degradation model; For rotating machinery The root mean square value of the vibration signal at time t; For the nonlinear observation function of the health state; For rotating machinery Rotational speed at any given moment; To observe noise; For neural networks; for The neural network parameters at time step; is the activation function for the first layer of the neural network; is the activation function for the second layer of the neural network; For the first layer of the neural network in The weight of each moment; For the second layer of the neural network in The weight of each moment; For the first layer of the neural network in Time offset; For the second layer of the neural network in The time offset.

[0012] In one possible implementation of the first aspect, the dimensionality reduction processing of the neural network parameters and the construction of a real-time anomaly detection index based on the dimensionality reduction result specifically includes: The neural network parameters are reduced to a preset dimension using principal component analysis, and the first principal component in the dimensionality reduction result is selected as the real-time anomaly detection index.

[0013] In one possible implementation of the first aspect, the real-time anomaly detection of the rotating machinery using statistical confidence criteria specifically includes: Based on the anomaly detection indicators of the rotating machinery mentioned before the current moment, the health threshold is determined by the 3Sigma criterion; If the abnormal detection index at the current moment exceeds the health threshold, the rotating machinery is determined to be abnormal.

[0014] In one possible implementation of the first aspect, the step of selecting the best degenerate fitting model from a plurality of preset degenerate fitting models based on the health status at the current time and before the current time, according to a model selection criterion, specifically includes: The model selection criterion is the BIC criterion; Collect health status data for the current moment and all moments prior to the current moment; The health status data are substituted into each of the preset degradation fitting models for fitting, the BIC value corresponding to each degradation fitting model is calculated, and the degradation fitting model with the smallest BIC value is selected as the best degradation fitting model.

[0015] In one possible implementation of the first aspect, the probability density distribution of the remaining service life of the rotating machinery is predicted using numerical simulation techniques based on the two-stage hybrid state-space model after updating the state transition coefficients. The specific process is as follows: The numerical simulation technology mentioned is Monte Carlo numerical simulation technology; Based on the two-stage hybrid state-space model after updating the state transition coefficients, the complete degradation process of the rotating machinery from its current healthy state to reaching the failure threshold is simulated by a preset number of Monte Carlo samplings, resulting in multiple healthy state degradation trajectories. The remaining service life corresponding to each degradation trajectory is calculated, and statistical analysis is performed on all remaining service life values ​​to generate the probability density distribution of the remaining service life of the rotating machinery.

[0016] In a possible implementation manner of the first aspect, the nonlinear filtering algorithm is an extended Kalman filtering algorithm, and the calculation process includes two steps of state prediction and state updating. In the state prediction step, the state prior value and the variance prior value at the current time are predicted based on the state posterior value and the variance posterior value at the previous time step and by using a state transition coefficient. In the state updating step, a Kalman filtering gain is calculated first, and then the state prior value and the variance prior value at the current time are updated by using the Kalman filtering gain to obtain the state posterior value and the variance posterior value at the current time as the final calculation result.

[0017] According to a second aspect of the present application, a computer device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the method for integrated detection of abnormal conditions and prediction of remaining service life of a rotating machine under time-varying working conditions when executing the computer program.

[0018] According to a third aspect of the present application, a computer readable storage medium is provided, which stores a computer program, and the computer program implements the method for integrated detection of abnormal conditions and prediction of remaining service life of a rotating machine under time-varying working conditions when executed by a processor.

[0019] According to a fourth aspect of the present application, a computer program product is provided, which implements the method for integrated detection of abnormal conditions and prediction of remaining service life of a rotating machine under time-varying working conditions when executed by a processor.

[0020] Compared with the prior art, the present application has at least the following beneficial effects: The application provides a rotating machinery time-varying working condition abnormality detection and residual useful life prediction integrated method, which solves the defects that the prior art can only complete abnormality identification, cannot track dynamic changes of the health state and quantize RUL, forms a closed loop of abnormality early warning, health state tracking and life quantization, simplifies the overall architecture of the PHM system, avoids error accumulation caused by multi-model switching, and can provide maintenance personnel with complete decision basis from whether to fail to how long to operate.

[0021] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the following preferred embodiments are described in detail below, and the accompanying drawings are used for description. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the specific embodiments of the present application, the following will briefly introduce the drawings needed to be used in the description of the specific embodiments. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without any creative labor on the basis of these drawings.

[0023] Figure 1 A flow chart of a rotating machinery time-varying working condition anomaly detection and remaining useful life prediction integrated method of the present application; Figure 2 A full life experiment table of the embodiment rolling bearing; Figure 3 The full life experiment signals of the embodiment three rolling bearings: (a) rotation speed signal; (b) vibration signal; (c) vibration signal RMS value; (d) rolling bearing failure type; Figure 4 The anomaly detection results of the embodiment: (a) neural network parameter change curve; (b) health index change curve; (c) vibration signal RMS value change curve; Figure 5 The failure time prediction results of the embodiment: (a) health state evaluation and failure time prediction; (b) best degradation fitting model selection results of degradation process; (c) working condition change curve; Figure 6 The RUL prediction results of the embodiment. DETAILED DESCRIPTION

[0024] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions of the present application will be described below in connection with the drawings, which apparently described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0025] As shown in Figure 1 The present application provides a rotating machinery time-varying working condition anomaly detection and remaining useful life prediction integrated method. In actual industrial scenarios, the running data of the rotating machinery can be obtained in real time by a data acquisition system deployed at the edge of the equipment, and an embedded processor or an industrial computer is used to execute each step of the method. The specific steps are as follows: S1, obtaining the vibration signal root mean square value and rotation speed in real time during the running process of the rotating machinery.

[0026] For example, the rotating machinery is a rolling bearing. Specifically, the original vibration signal during the running process of the rolling bearing is collected by a vibration sensor (such as an acceleration sensor) installed on the bearing seat or adjacent structure, and the real-time rotation speed of the bearing is obtained synchronously by a device control system or a rotation speed sensor. After the vibration signal is input to the data acquisition card for analog-to-digital conversion through the signal conditioning circuit (including amplification, filtering, etc.), the root mean square value (RMS) of the vibration signal is calculated. The collected vibration signal root mean square value and rotation speed value are stored in time sequence.

[0027] S2. Initialize a two-stage hybrid state-space model. The two-stage hybrid state-space model is used to characterize the functional relationship between rotational speed, health state, and root mean square value of vibration signal during the entire life cycle of rotating machinery. It includes a state equation and an observation equation. The state transition coefficient in the state equation determines whether the two-stage hybrid state-space model is in a healthy stage or a degenerate stage. The observation equation uses a neural network as a nonlinear observation function.

[0028] Specifically, in this embodiment, the two-stage hybrid state-space model takes the following form: Equations of state:

[0029] Observation equation:

[0030] in,

[0031]

[0032] In the formula: For rotating machinery Constant health status; For rotating machinery Constant health status; For rotating machinery State transition coefficient at time t; State noise; The best-fit model for degradation; and These are the parameters in the best-fit degradation model; For rotating machinery The root mean square value of the vibration signal at time t; For the nonlinear observation function of the health state; For rotating machinery Rotational speed at any given moment; To observe noise; For neural networks; for The neural network parameters at time step; is the activation function for the first layer of the neural network; is the activation function for the second layer of the neural network; For the first layer of the neural network in The weight of each moment; For the second layer of the neural network in The weight of each moment; For the first layer of the neural network in Time offset; For the second layer of the neural network in The time offset.

[0033] For example, the neural network uses a shallow neural network with a 3-10-1 architecture (i.e., 3 inputs, 10 intermediate computing nodes, and 1 output) to approximate the observation function. The activation function of the neural network can be the Sigmoid function or the ReLU function.

[0034] Initialize the two-stage hybrid state-space model, that is, initialize the state noise and observation noise.

[0035] S3. In the healthy phase, the state transition coefficient is set to 1, and the root mean square value of the vibration signal and the rotation speed at the current moment are input into the two-stage hybrid state space model. The neural network parameters at the current moment are calculated using a nonlinear filtering algorithm.

[0036] In other words, during the healthy phase, the state transition coefficient is fixed at 1, indicating that the healthy state has no degradation. The two-stage hybrid state-space model at this time is:

[0037] It should be noted that during the entire life cycle of rotating machinery, the current health status of the rotating machinery and the current neural network parameters need to be updated and solved alternately, as follows:

[0038] in, For rotating machinery The root mean square value of the predicted vibration signal at time; For rotating machinery Posterior value of health status at any given time; For rotating machinery The posterior values ​​of the neural network parameters at time step 1; For rotating machinery Prior values ​​of neural network parameters at time step; In one possible implementation, the nonlinear filtering algorithm is the extended Kalman filter algorithm.

[0039] The process of calculating the current health state of rotating machinery includes two steps: state prediction and state update. In the state prediction step, based on the posterior state value and variance posterior value of the previous time step, the state prior value and variance prior value of the current time step are predicted using the state transition coefficients, as follows:

[0040]

[0041] in, For rotating machinery Prior values ​​of health status at any given time; For rotating machinery Posterior value of health status at any given time; For rotating machinery Prior values ​​of health status covariance at time t; For rotating machinery Posterior value of health status covariance at time 1; for Health status noise covariance at any given time.

[0042] In the state update step, the Kalman filter gain is first calculated, and then the state prior value and variance prior value at the current time are updated using the Kalman filter gain to obtain the state posterior value and variance posterior value at the current time, which are used as the final calculation results, as follows: The Kalman filter gain is expressed as follows:

[0043] in, For health status Kalman gain at time step; for The noise covariance of the health status at any given time; The Jacobian matrix, representing the nonlinear observation function of health status, can be expressed as follows:

[0044] in, For rotating machinery Rotational speed at any given moment; for Prior values ​​of neural network parameters at time step; Then, the state prior value and variance prior value are updated using the Kalman filter gain, as shown below:

[0045]

[0046] in, For rotating machinery Posterior value of health status at any given time; Used to fit the observation function The update process of neural network parameters can be represented by the following state-space model:

[0047] in, for The neural network parameters at time step; It is the identity matrix; a nonlinear observation function of the neural network parameter; Similarly, the process of calculating the neural network parameter at the current time also includes two steps of state prediction and state update: In the state prediction step, the posterior value of the NN parameter vector and the covariance matrix is represented as:

[0048]

[0049] wherein, is the prior value of the neural network parameter of the rotating machinery at time t; is the posterior value of the neural network parameter of the rotating machinery at time t; is the prior value of the neural network parameter covariance of the rotating machinery at time t; is the posterior value of the neural network parameter covariance of the rotating machinery at time t; is the noise covariance of the neural network parameter at time t; is the prior value of the neural network parameter covariance of the rotating machinery at time t; is the posterior value of the neural network parameter covariance of the rotating machinery at time t; is the noise covariance of the neural network parameter at time t; In the state update step, the Kalman filter gain can be represented as follows: wherein, is the Kalman gain of the neural network parameter at time t;

[0050] is the observation noise covariance of the neural network parameter at time t; and is the Jacobian matrix of the nonlinear observation function of the neural network parameter; is the Jacobian matrix of the nonlinear observation function of the neural network parameter; is the Jacobian matrix of the nonlinear observation function of the neural network parameter; is the Jacobian matrix of the nonlinear observation function of the neural network parameter;

[0051] wherein, represents a small perturbation.

[0052] Finally, the neural network parameter vector and the covariance matrix are updated as follows:

[0053]

[0054] S4, performing dimensionality reduction processing on the neural network parameter, constructing a real-time anomaly detection index based on the dimensionality reduction result, and combining a statistical confidence criterion to perform real-time anomaly detection on the rotating machinery.

[0055] ​​In an implementation, the neural network parameters are reduced in dimension to a preset dimension using a principal component analysis (PCA) algorithm, and a first principal component in the reduced dimension result is selected as the real-time anomaly detection index.

[0056] For example, the preset dimension is 3.

[0057] In an implementation, the statistical confidence criterion is specifically a 3Sigma criterion, and the real-time anomaly detection of the rotating machinery is specifically as follows: Based on the anomaly detection index of the rotating machinery before the current time, a health threshold is determined by the 3Sigma criterion; If the anomaly detection index at the current time exceeds the health threshold, it is determined that the rotating machinery is abnormal; otherwise, the rotating machinery is in a healthy state.

[0058] S5, if no anomaly is detected, returning to S3 to continue the health stage processing; if an anomaly is detected, entering S6.

[0059] S6, in the degradation stage, the root mean square value of the vibration signal at the current time and the rotating speed 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 time.

[0060] In this embodiment, the calculation process of the nonlinear filtering algorithm is the same as step S3, which will not be repeated here.

[0061] S7, based on the health state at the current time and before the current time, a best degradation fitting model is selected from a plurality of preset degradation fitting models based on a model selection criterion, and the state transition coefficient is updated based on the best degradation fitting model.

[0062] In an implementation, the model selection criterion uses a BIC criterion, and the specific implementation process is as follows: collecting health state data at the current time and all times before the current time ; the health state data is respectively substituted into each preset degradation fitting model for fitting, the BIC value corresponding to each degradation fitting model is calculated, and the degradation fitting model with the minimum BIC value is selected as the best degradation fitting model.

[0063] In this embodiment, the plurality of preset degradation fitting models include four common health state degradation modes, which are as follows: A linear degradation fitting model, that is ; A polynomial degradation fitting model, that is ; A logarithmic degradation fitting model, that is ); An exponential degradation fitting model, that is ).

[0064] S8, based on the two-stage hybrid state space model after updating the state transition coefficient, the probability density distribution of the remaining useful life of the rotating machinery is predicted by using a numerical simulation technique.

[0065] In an implementation manner, the numerical simulation technique adopts a Monte Carlo numerical simulation technique, and the specific implementation process is as follows: based on the two-stage hybrid state space model after updating the state transition coefficient, a complete degradation process of the rotating machinery from the current health state to reaching the failure threshold is simulated through a preset number of Monte Carlo samplings to obtain a plurality of health state degradation trajectories; the corresponding remaining useful life is calculated based on each degradation trajectory, and statistical analysis is performed on all the remaining useful life values to generate the probability density distribution of the remaining useful life of the rotating machinery. The final RUL prediction result is quantitatively evaluated by using an absolute error (ARE), and the calculation formula is as follows:

[0066] wherein, is the predicted failure time of the rotating machinery; is the actual failure time of the rotating machinery.

[0067] For example, the number of Monte Carlo samplings is 100.

[0068] Next, the technical solutions described in the present application will be described in detail with the full-life data set collected on a rolling bearing full-life test bench as an example. As shown in Figure 2 , the rolling bearing accelerated life test bench mainly consists of an alternating current motor, an optical speed sensor, two supporting bearings and two hydraulic loading systems. The experimental bearing used in the experiment is an ER-16K deep groove ball bearing.

[0069] Step 1: Real-time acquisition of the root mean square value of the vibration signal and the rotating speed of the rolling bearing during operation.

[0070] During the experiment, the hydraulic loading system is used to apply axial and radial loads, and the accelerometer is used to collect the original vibration signal of the bearing, and the optical speed sensor is used to synchronously acquire the real-time rotating speed of the bearing. The sampling frequency is set to 20480 Hz, and the data is collected every 10 minutes, and the collection time is 3 seconds each time. The root mean square value (RMS) of the collected original vibration signal is calculated, and the value is stored in time sequence together with the rotating speed value. Figure 3 (b1-b3) are the collected vibration signals; Figure 3 (c1-c3) are the vibration signal RMS value change curves; Figure 3 (b1-b3) are the final failure types of the rolling bearing; In order to simulate the time-varying working condition, two types of working conditions are designed in this experiment: Case 1 (short time scale periodic variation): the rotating speed varies at equal intervals between 10 Hz and 50 Hz, and changes once every 0.5 hour, which is approximately continuous and periodic, as shown in Figure 3 (a1-a2).

[0071] Case 2 (long time scale random variation): the rotating speed shows significant uncertainty, rather than periodic variation at equal intervals, and changes randomly once every 10 hours, as shown in Figure 3 (a3).

[0072] A total of three experiments were conducted to verify the effectiveness of the method. Step 2, initialize the two-stage hybrid state space model.

[0073] It should be noted that the two-stage hybrid state space model is constructed based on the ordinary state space model to describe the functional relationship between the rotating speed, health state and vibration signal root mean square value during the whole life cycle of the rolling bearing under time-varying working conditions. The health state is initialized to 0; the neural network parameters are all initialized to 0.2; the health state error covariance is initialized to 0.01; and the neural network parameter covariance is initialized to the unit matrix .

[0074] The two-stage hybrid state space model is composed of a state equation and an observation equation.

[0075] The state equation is . In the health stage, the state transition coefficient is set to 1. In the degradation stage, the state transition coefficient is determined by the optimal degradation fitting model.

[0076] The observation equation is , and a shallow neural network with a 3-10-1 architecture is used as the nonlinear observation function. The 3 inputs of the shallow neural network are: the health state at the current time, the rotating speed at the current time, and the product of the two.

[0077] Step 3, in the health stage, set the state transition coefficient to 1, and input the vibration signal root mean square value and the rotating speed at the current time into the two-stage hybrid state space model, and use the extended Kalman filter algorithm to calculate the neural network parameters at the current time.

[0078] In this embodiment, the 3-10-1 neural network structure has 51-dimensional parameters, and the variation curves of each dimension are shown in Figure 4 (a1-a3). In the health stage, the neural network parameters gradually converge to a constant value. After the initial failure occurs, the neural network parameters begin to diverge rapidly.

[0079] Step 4, dimension reduction is performed on the neural network parameters, a real-time abnormality detection index is constructed based on the dimension reduction result, and a statistical confidence criterion is combined to perform real-time abnormality detection on the rolling bearing.

[0080] The principal component analysis (PCA) algorithm is used to reduce the 51-dimensional neural network parameter vector to 3 dimensions, and the first principal component in the dimension reduction result is selected as the real-time abnormality detection index.

[0081] Based on the abnormality detection index before the current time, the health threshold is determined through the 3Sigma criterion. If the abnormality detection index at the current time exceeds the health threshold, it is determined that the rolling bearing is abnormal; otherwise, the bearing is in a healthy state. As shown in Figure 4 (b1-b3) shows that the initial abnormal points of the three test bearings are the 318th, 1403rd and 1970th data samples, respectively. Before the occurrence of the abnormal point, the detection index remains basically stable, and after the occurrence of the abnormal point, the detection index diverges rapidly.

[0082] Figure 4 (c1-c3) is the change curve of the RMS value of the vibration signal, which is less stable and less effective in abnormality detection than the health index constructed by the method, proving the effectiveness and superiority of the constructed index.

[0083] Step 5, if no abnormality is detected, return to S3 and continue the health stage processing; if abnormality is detected, proceed to S6.

[0084] Step 6, in the degradation stage, the root mean square value of the vibration signal at the current time and the rotational speed are input into the two-stage hybrid state space model, and the extended Kalman filter algorithm is used to calculate the health state of the rolling bearing at the current time.

[0085] Step 7, based on the health state at the current time and before the current time, the BIC criterion is used to select the best degradation fitting model from a plurality of preset degradation fitting models, and the state transition coefficient is updated based on the best degradation fitting model.

[0086] Step 8, based on the two-stage hybrid state space model after updating the state transition coefficient, the Monte Carlo numerical simulation technique is used to predict the probability density distribution of the remaining useful life of the rolling bearing.

[0087] After detecting the abnormality, the rolling bearing enters the degradation stage. The Monte Carlo sampling number is set to 100, the Monte Carlo numerical simulation technique and the degradation stage hybrid state space model constructed in step 3 are used to simulate the health state degradation trajectory, and thus the RUL probability density function is obtained. Figure 5 The real-time updating process of the health state and the optimal degradation fitting model used in the bearing degradation process are given. As shown in Figure 5(a1-a3) shows that the predicted health state alleviates the impact of time-varying working conditions to some extent and presents a monotonically increasing trend in general. The figure further gives the prediction results of failure time probability distribution at three time points and the confidence interval of the predicted health state trajectory. From these prediction results, it can be seen that the predicted failure time probability distribution either completely covers the actual failure time or gradually converges to the range of the true failure time. This is because as the data accumulates, the optimal degradation fitting model selected by the algorithm will become more accurate, resulting in more reliable and accurate prediction results. Figure 5 (b1-b3) are the optimal degradation fitting models adaptively selected by the method in the degradation process of three rolling bearings. Figure 5 (c1-c3) are the running speed change curves of three rolling bearings. Figure 6 (a1-a3) are the RUL prediction results, and the RUL prediction results of the method of the present application all fluctuate around the actual RUL in the figure. The quantitative evaluation indexes in Table 1 further show the accuracy of the RUL prediction of the method.

[0088] Table 1

[0089] In another embodiment of the present application, a computer device is provided, which comprises a processor and a memory, the memory is used to store a computer program, the computer program comprises program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, and are suitable for implementing one or more instructions, and are specifically suitable for loading and executing one or more instructions in the computer storage medium to implement a corresponding method flow or a corresponding function; the processor of the embodiment of the present application can be used for the operation of the rotating machinery time-varying working condition anomaly detection and remaining useful life prediction integrated method.

[0090] In still another embodiment of the present application, the present application also provides a storage medium, specifically a computer readable storage medium (Memory), which is a memory device in a computer device, used for storing programs and data. It can be understood that the computer readable storage medium here can include the built-in storage medium in the computer device, and of course can also include the extended storage medium supported by the computer device. The computer readable storage medium provides a storage space, which stores the operating system of the terminal. Moreover, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer readable storage medium here can be a high-speed RAM memory, or a non-volatile memory such as at least one disk memory. One or more instructions stored in the computer readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the above-mentioned embodiment about the rotating machinery time-varying working condition anomaly detection and residual service life prediction integrated method.

[0091] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0092] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts 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, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device that implements the flowcharts and / or block diagrams. Figure 1 The function specified in one flow or multiple flows and / or blocks Figure 1 The device that implements the function specified in one flow or multiple flows and / or blocks.

[0093] These computer program instructions can also be stored in a computer readable memory capable of directing the computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce a product including instruction devices, which implement the flowcharts and / or block diagrams. Figure 1one or more processes and / or blocks Figure 1 the function specified in the one or more blocks.

[0094] These computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable data processing devices to generate a computer-implemented process, so that the instructions executed on the computer or other programmable data processing devices provide a process for implementing the flow Figure 1 one or more processes and / or blocks Figure 1 Figure 1 the steps of the function specified in the one or more blocks.

[0095] The present application also provides a computer program product, since the computer program product is used to execute any one of the above-mentioned rotating machinery time-varying working condition anomaly detection and residual service life prediction integrated method. Since the computer program product provided by the present application belongs to the same inventive concept as the above-mentioned rotating machinery time-varying working condition anomaly detection and residual service life prediction integrated method, the computer program product provided by the present application has all the advantages of the above-mentioned rotating machinery time-varying working condition anomaly detection and residual service life prediction integrated method, so the beneficial effects of the computer program product provided by the present application will not be described one by one.

[0096] In the present application, the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, different embodiments or examples described in the present specification and the features of different embodiments or examples can be combined and combined by those skilled in the art without contradiction.

[0097] Finally, it should be noted that: the above-described embodiments are only specific embodiments of the present application, which are used to illustrate the technical solutions of the present application, but not to limit them, the protection scope of the present application is not limited thereto, although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand: any person skilled in the art within the technical range disclosed by the present application, still can modify or easily think of changes to the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, all should be covered in the protection scope of the present application.

Claims

1. A method for integrated abnormality detection and remaining useful life prediction of time-varying working conditions of rotating machinery, characterized in that, The method comprises the following steps: S1, acquiring the root mean square value of a vibration signal and a rotating speed in a running process of a rotating machine in real time; S2, initializing a two-stage hybrid state space model, the two-stage hybrid state space model being used to represent a functional relationship among the rotating speed, a health state and the root mean square value of the vibration signal in a whole life cycle of the rotating machine, comprising a state equation and an observation equation, wherein a state transition coefficient in the state equation determines whether the two-stage hybrid state space model is in a health stage or a degradation stage, and the observation equation adopts a neural network as a nonlinear observation function; S3, in the health stage, setting the state transition coefficient as 1, inputting the root mean square value of the vibration signal and the rotating speed at a current time into the two-stage hybrid state space model, and calculating neural network parameters at the current time by using a nonlinear filtering algorithm; S4, performing dimension reduction processing on the neural network parameters, constructing a real-time anomaly detection index based on a dimension reduction result, and performing real-time anomaly detection on the rotating machine in combination with a statistical confidence criterion; S5, if no anomaly is detected, returning to S3 to continue the health stage processing; if an anomaly is detected, entering S6; S6, in the degradation stage, inputting the root mean square value of the vibration signal and the rotating speed at the current time into the two-stage hybrid state space model, and calculating the health state of the rotating machine at the current time by using the nonlinear filtering algorithm; S7, selecting an optimal degradation fitting model from a plurality of preset degradation fitting models based on a model selection criterion according to the health state at the current time and before the current time, and updating the state transition coefficient based on the optimal degradation fitting model; S8, predicting a probability density distribution of a remaining service life of the rotating machine by using a numerical simulation technology based on the two-stage hybrid state space model after the state transition coefficient is updated.

2. The integrated method for abnormality detection and remaining useful life prediction of time-varying working conditions of rotating machinery according to claim 1, characterized in that, The specific form of the two-stage hybrid state space model is: State equation: Observation equation: wherein, In the formula: is the health state of the rotating machinery at time ; is the health state of the rotating machinery at time ; is the state transition coefficient of the rotating machinery at time ; is the state noise; is the optimal degradation fitting model; and are parameters in the optimal degradation fitting model; is the root mean square value of the vibration signal of the rotating machinery at time ; is the health state nonlinear observation function; is the rotating speed of the rotating machinery at time ; is the observation noise; is the neural network; is the neural network parameter at time ; is the activation function of the first layer neural network; is the activation function of the second layer neural network; is the weight of the first layer neural network at time ; is the weight of the second layer neural network at time ; is the bias of the first layer neural network at time ; is the bias of the second layer neural network at time .

3. The integrated method for abnormality detection and remaining useful life prediction of time-varying working conditions of rotating machinery according to claim 1, characterized in that, The dimension reduction processing on the neural network parameters and the construction of the real-time anomaly detection index based on the dimension reduction result are specifically as follows: the neural network parameters are reduced to a preset dimension by using a principal component analysis algorithm, and a first principal component in the dimension reduction result is selected as the real-time anomaly detection index.

4. The integrated method for abnormality detection and remaining useful life prediction of time-varying working conditions of rotating machinery according to claim 1, characterized in that, The real-time anomaly detection on the rotating machine in combination with the statistical confidence criterion is specifically as follows: a health threshold is determined by using a 3Sigma criterion based on the anomaly detection index of the rotating machine before the current time; if the anomaly detection index at the current time exceeds the health threshold, it is determined that the rotating machine has an anomaly.

5. The integrated method of abnormality detection and remaining useful life prediction for time-varying working conditions of a rotating machine according to claim 1, characterized in that, The selection of the optimal degradation fitting model from the plurality of preset degradation fitting models based on the model selection criterion according to the health state at the current time and before the current time is specifically as follows: the model selection criterion is a BIC criterion; health state data at the current time and all times before the current time are collected; the health state data are respectively substituted into each preset degradation fitting model to perform fitting, a BIC value corresponding to each degradation fitting model is calculated, and a degradation fitting model with the minimum BIC value is selected as the optimal degradation fitting model.

6. The integrated method of abnormality detection and remaining useful life prediction for time-varying working conditions of a rotating machine according to claim 1, characterized in that, The two-stage hybrid state space model based on the updated state transition coefficient adopts a numerical simulation technique to predict the probability density distribution of the remaining useful life of the rotating machinery, and the specific process is as follows: The numerical simulation technique is a Monte Carlo numerical simulation technique. Based on the two-stage hybrid state space model based on the updated state transition coefficient, the complete degradation process of the rotating machinery from the current health state to the failure threshold is simulated through a preset number of Monte Carlo samples, and a plurality of health state degradation trajectories are obtained. Based on each degradation trajectory, the corresponding remaining useful life is calculated, and statistical analysis is performed on all the remaining useful life values to generate the probability density distribution of the remaining useful life of the rotating machinery.

7. The integrated method of abnormality detection and remaining useful life prediction for time-varying working conditions of a rotating machine according to claim 1, characterized in that, The nonlinear filtering algorithm is an extended Kalman filtering algorithm, and the calculation process includes two steps of state prediction and state update: In the state prediction step, based on the state posterior value and the variance posterior value of the previous time step, the state prior value and the variance prior value at the current time are predicted using the state transition coefficient. In the state update step, the Kalman filter gain is first calculated, and then the Kalman filter gain is used to update the state prior value and the variance prior value at the current time to obtain the state posterior value and the variance posterior value at the current time as the final calculation result.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to realize the rotating machinery time-varying working condition anomaly detection and remaining useful life prediction integrated method according to any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 8. The computer program is executed by the processor to realize the rotating machinery time-varying working condition anomaly detection and remaining useful life prediction integrated method according to any one of claims 1 to 7.

10. A computer program product, characterised in that, The computer program product is executed by the processor to realize the rotating machinery time-varying working condition anomaly detection and remaining useful life prediction integrated method according to any one of claims 1 to 7.

Images