Dynamic reliability evaluation method and system for gear box of large road maintenance machine

Abstract

The invention discloses a dynamic reliability evaluation method for a gear box of a large road maintenance machine. The method comprises the following steps: S1, multi-source data acquisition and feature extraction; s2, establishing Bayesian network-fault mode mapping based on the extracted features; s3, establishing a non-uniform Markov chain model corresponding to the health state of the mechanical gearbox based on a dynamic modeling method; and S4, performing residual life dynamic prediction and decision optimization of the mechanical gearbox based on the Bayesian network-fault mode mapping and the non-uniform Markov chain model. The invention further discloses a corresponding system, electronic equipment and a computer readable storage medium.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance and condition monitoring of large track maintenance machinery, and particularly relates to a method and system for dynamically evaluating the reliability of a gearbox of large track maintenance machinery. Background Art

[0002] The gearbox of large track maintenance machinery has been bearing high-frequency vibration, impact load and complex environmental stress for a long time. Its degradation process presents non-linear, time-varying and multi-modal coupling characteristics. However, after retrieval, there is no existing method and system for dynamically evaluating the reliability of the gearbox of large track maintenance machinery in the prior art. The existing reliability evaluation methods for mechanical gearboxes have the following problems: 1. The traditional threshold alarm method cannot capture the progressive degradation characteristics of the gearbox and ignores the influence of working condition fluctuations on the health state; 2. There is a lack of an effective fusion mechanism for multi-dimensional data such as oil, temperature, and vibration; 3. Existing methods mostly rely on after-sales maintenance and lack the ability to dynamically quantify the remaining service life and evaluate the confidence interval. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for dynamically evaluating the reliability of a gearbox of large track maintenance machinery based on a Bayesian network and a non-uniform Markov chain, which is a method and system for dynamically evaluating the reliability of a gearbox of large track maintenance machinery that integrates multi-source sensing data, Bayesian network reasoning and non-uniform Markov chain modeling.

[0004] The first aspect of the present invention lies in providing a method for dynamically evaluating the reliability of a gearbox of large track maintenance machinery, including the following steps: S1, multi-source data acquisition and feature extraction; S2, establishing a Bayesian network-fault mode mapping based on the extracted features; S3, establishing a non-uniform Markov chain model corresponding to the health state of the mechanical gearbox based on a dynamic modeling method; S4, performing dynamic prediction of the remaining life and decision optimization of the mechanical gearbox based on the Bayesian network-fault mode mapping and the non-uniform Markov chain model.

[0005] Preferably, the S1 includes: S11, performing multi-source data acquisition, including: deploying vibration acceleration sensors, temperature sensors and oil particle counters, and collecting the vibration time-domain waveform of the gearbox, the gearbox temperature and the concentration of lubricating oil wear particles under different operating conditions based on the deployed vibration acceleration sensors, temperature sensors and oil particle counters; at the same time, recording the actual health state change of the mechanical gearbox. S12. Extract the features of multi-source data, including: performing wavelet decomposition on the vibration time-domain waveform of the gearbox and extracting the eigenvalue as the frequency-domain degradation index.

[0006] Preferably, the S2 includes: S21. Establish the network topology structure diagram of the Bayesian network. The network topology structure diagram of the Bayesian network includes three layers of defined nodes. Each layer of nodes includes a root node and multiple leaf nodes. The root node is used to represent the fault mode; the multiple leaf nodes are used to represent the health indicators and operating conditions. Among them, the fault modes include: F1: tooth surface pitting, F2: bearing spalling, F3: cage damage; the health indicators include: H1: vibration, H2: temperature, H3: oil particles; the operating conditions include: C1: load rate, C2: ambient temperature. S22. Calculate multiple conditional probability tables, including: based on the historical fault case database, learning the conditional probability relationship between multiple root nodes and the multiple leaf nodes through a data mining algorithm, so as to form the multiple conditional probability tables. S23. Establish a fault-feature-condition causal inference network based on the network topology structure diagram and the multiple conditional probability tables. The fault-feature-condition causal inference network contains the Bayesian network-fault mode mapping.

[0007] Preferably, the S3 includes: S31. Determine the state space, divide the health state of the gearbox into {normal, mild degradation, moderate degradation, severe fault}, and represent them by S 1 、S 2 、S 3 、S 4 respectively, so as to determine the state space {S 1 ,S 2 ,S 3 ,S 4}; S32. Obtain the collected multi-source data. S33. Divide the time interval, including: selecting an appropriate time interval Δt according to the operating characteristics of the mechanical gearbox and the data acquisition frequency. S34. Determine the transition probability, including: calculating the probability P i of transferring from state S j to state S ij at different time intervals nΔt according to historical data, that is: P ij (n)=P(X (k+n)Δt =S j |X kΔt =S i ); S35. Establish a non - homogeneous Markov chain model corresponding to the health state of the mechanical gearbox, including: representing the transition probability matrix P(n) of the non - homogeneous Markov chain model in matrix form, and its elements are P ij (n); assuming that at time kΔt, the probability that the gearbox is in state S i is π i (k), then the probability π j of being in state S j at time (k + n)Δt can be calculated by: π j (k + n)=sum(π i (k)P ij (n)) (where i = 1, 2, 3, 4); S36. Conduct verification and optimization of the non - homogeneous Markov chain model.

[0008] Preferably, the S36 includes: (1) Use new experimental data to verify the established non - homogeneous Markov chain model, and evaluate the accuracy of the model by comparing the index differences between the model prediction results and the actual observation results; among them, the index differences include mean - square error index differences; (2) If the non - homogeneous Markov chain model is inaccurate, analyze the reasons and optimize the non - homogeneous Markov chain model; among them, the optimization includes: readjusting the state space and / or improving the calculation method of the transition probability.

[0009] Preferably, the S4 includes: S41. Based on the Viterbi algorithm, backtrack the optimal state path, calculate the expected time from the current state to the failure threshold and the 95% confidence interval of the remaining life; S42. Input the real - time data stream after feature extraction into the Bayesian network, and output the probability of fault occurrence based on the Bayesian network - fault mode mapping; S43. Drive the state transition of the non - homogeneous Markov chain model based on the probability of fault occurrence and the working condition parameters of the mechanical gearbox, and update the prediction result of the remaining life of the mechanical gearbox; S44. Trigger an early warning or generate a maintenance suggestion based on the comparison between the 95% confidence interval of the remaining life and the maintenance economy model, and provide decision - making optimization according to the state of the mechanical gearbox.

[0010] Preferably, the backtracking of the optimal state path in the S41 based on the Viterbi algorithm includes: (1) Initialization: Based on the known non - homogeneous Markov chain model, set the state probability distribution π 0 at the initial moment, that is, the probability that the gearbox is in each state at the initial moment; For each state \(i\), set the initial path metric \(\delta\) 0 (i)=\(\pi\) 0 (i); And record the predecessor state \(\psi\) of the initial state 0 (i) = 0; (2) Perform recursive calculations: Starting from the initial moment, at each time step \(t\), for each state \(j\), calculate the path metric \(\delta\) for the transition from all possible previous states \(i\) to state \(j\) t (j)=\(\max\) i [\(\delta\) t-1 (i)P ij (t)]; Meanwhile, record the predecessor state \(\psi\) that maximizes this path metric t (j)=\(\arg\max\) i [\(\delta\) t-1 (i)P ij (t)], where P ij (t) is the transition probability from state \(i\) to state \(j\) at time \(t\); (3) Determine the termination conditions, including: Stop the calculation when reaching a certain termination time or reaching the state corresponding to the failure threshold. Assume that the termination condition is reached at time \(T\), and find the state \(i\) that maximizes \(\delta\) T (i) * , and this state is the final optimal state; (4) Backtrack the path: Starting from the final optimal state \(i\) * and using the recorded predecessor state information \(\psi\) t (i), backtrack to obtain the optimal state path \(\{i\) * T , \(i\) * T-1 ,… \(i\) * 0 \}.

[0011] Preferably, the calculation of the expected time from the current state to the failure threshold in S41 includes: (1) Determine the transition time probability distribution: According to the non-uniform Markov chain model, determine the time probability distribution required to transfer from each state to other states, and fit the probability density function of the state transition time through historical data; (2) Expected time calculation: Starting from the current state, along the optimal state path, calculate the expected time of each state transition according to the state transition time probability distribution; Let the expected time from state \(i\) to state \(j\) be \(E[t\) ij , along the optimal state path \(\{i\) * T , \(i\) * T-1 ,… \(i\)* 0}, the current state (assumed to be i * 0 ) to the failure threshold (assumed that the final state is i * T ) the expected time E[T]=sumE[t i*ti*t+1 (t = 0,..., T).

[0012] Preferably, the calculation of the 95% confidence interval of the remaining life in S41 includes: (1) Simulation sampling: Using the non-uniform Markov chain model and the state transition time probability distribution, a large number of simulations (such as N times) are carried out. Each simulation starts from the current state and calculates the time T to reach the failure threshold along the randomly generated state path (k) , where k = 1, 2,..., N; (2) Sorting and confidence interval calculation: Sort the times T obtained from N simulations (k) in ascending order; According to the definition of the confidence interval, the lower limit of the 95% confidence interval is the 0.025Nth value after sorting, and the upper limit is the 0.975Nth value; that is, the 95% confidence interval of the remaining life is [T lower , T upper , where T lower and T upper are the calculated lower and upper limit values respectively.

[0013] The second aspect of the present invention provides a dynamic reliability evaluation system for large track maintenance machinery gearboxes for implementing the method of the first aspect. The system includes: A data acquisition and feature extraction module (101) for multi-source data acquisition and feature extraction; A Bayesian network establishment module (102) for establishing a Bayesian network - failure mode mapping based on the extracted features; A non-uniform Markov chain model establishment module (103) for establishing a non-uniform Markov chain model corresponding to the health state of the mechanical gearbox based on the dynamic modeling method; A reliability evaluation module (104) for performing dynamic prediction of the remaining life and decision optimization of the mechanical gearbox based on the Bayesian network - failure mode mapping and the non-uniform Markov chain model.

[0014] The third aspect of the present invention provides an electronic device, including a processor and a memory. The memory stores multiple instructions, and the processor is used to read the instructions and execute the method as described in the first aspect.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium storing a plurality of instructions that can be read and executed by a processor to perform the method as described in the first aspect.

[0016] Advantages of the method and system of the present invention:

[0017] At the same time, it realizes the fusion and index extraction of multi-source heterogeneous data (vibration, temperature, oil), the modeling of dynamic time-varying transition probability; as well as the confidence interval prediction of the remaining useful life and the recommendation of adaptive maintenance decisions; it solves the problems that the traditional threshold alarm method of the prior art cannot capture the progressive degradation characteristics of the gearbox and ignores the influence of working condition fluctuations on the health state; there is a lack of an effective fusion mechanism for multi-dimensional data such as oil, temperature, and vibration; existing methods mostly rely on after-the-fact maintenance and lack the ability to dynamically quantify the remaining useful life and evaluate the confidence interval. Description of the Drawings

[0018] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the related art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the related art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0019] Figure 1 It is a flowchart of a method for dynamically evaluating the reliability of a gearbox of a large track maintenance machine provided according to an embodiment of the present invention; Figure 2 It is an architecture of a system for dynamically evaluating the reliability of a gearbox of a large track maintenance machine provided according to an embodiment of the present invention; Figure 3 It is a structural diagram of an electronic device provided according to an embodiment of the present invention. Detailed Embodiments

[0020] The following will clearly and completely describe the technical solutions of the present invention with reference to the drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the protection scope of the present invention.

[0021] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. In addition, the terms "first", "second", "third" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance.

[0022] In the description of the present invention, it should be noted that unless otherwise clearly specified and limited, the terms "installed", "connected", "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0023] A gearbox dynamic reliability evaluation method and system based on a BN-NHMC fusion model aims to achieve: 1. Fusion and index extraction of multi-source heterogeneous data (vibration, temperature, oil); 2. Modeling of dynamic time-varying transition probability; 3. Prediction of the confidence interval of the remaining service life and recommendation of adaptive maintenance decisions. Embodiment 1

[0024] As Figure 1 shown, this embodiment provides a method for evaluating the dynamic reliability of a large track maintenance machine gearbox, including the following steps: S1. Multi-source data acquisition and feature extraction; As a preferred implementation manner, the S1 includes: S11. Perform multi-source data acquisition, including: deploying vibration acceleration sensors, temperature sensors, and oil particle counters, and collecting the vibration time-domain waveform of the gearbox, the gearbox temperature, and the concentration of lubricating oil wear particles under different operating conditions based on the deployed vibration acceleration sensors, temperature sensors, and oil particle counters; at the same time, record the actual health status change of the mechanical gearbox. S12. Perform feature extraction of multi-source data, including: performing wavelet decomposition on the vibration time-domain waveform of the gearbox and extracting feature values as frequency-domain degradation indicators.

[0025] S2. Establish a Bayesian network-fault mode mapping based on the extracted features; Bayesian Networks (BN) is a probabilistic network that uses a graphical approach for decision analysis. It is a model for representing and reasoning about uncertain factors based on probability analysis and graph theory, and is an information representation framework that combines causal relationships and probabilistic knowledge. The BN network is a directed acyclic graph (DAG) containing a conditional probability table. In the network topology structure diagram, nodes represent variables (or events), and the arcs between nodes (pointing from the cause event to the result event) represent the direct causal relationship between the parent node and the child node, and are annotated in the form of a two-dimensional conditional probability table (CPT). The probability of a node without any parent node is its prior probability. The node variables can be abstractions of any variables, such as the state of equipment components, test values, observed phenomena, etc. BN = <G, {CPT}>, where G represents the network topology structure diagram, and {CPT} represents the conditional probability tables associated with each node variable. A graph G and n conditional probability tables constitute a Bayesian network.

[0026] As a preferred embodiment, S2 includes: S21, establishing the network topology structure diagram of the Bayesian network. The network topology structure diagram of the Bayesian network includes three layers of defined nodes. Each layer of nodes includes a root node and multiple leaf nodes. The root node is used to represent the failure mode (in this embodiment, the failure modes include: F1: tooth surface pitting, F2: bearing spalling, F3: cage damage, etc.); the multiple leaf nodes are used to represent the health indicators (in this embodiment, the health indicators include: H1: vibration, H2: temperature, H3: oil particle) and operating condition parameters (in this embodiment, the operating condition parameters include: C1: load ratio, C2: ambient temperature).

[0027] S22, calculating multiple conditional probability tables, including: based on the historical failure case library, learning the conditional probability relationships between multiple root nodes and the multiple leaf nodes through a data mining algorithm, so as to form the multiple conditional probability tables; S23, establishing a failure - feature - operating condition causal inference network based on the network topology structure diagram and the multiple conditional probability tables. The failure - feature - operating condition causal inference network contains the Bayesian network - failure mode mapping.

[0028] S3, establishing a non - homogeneous Markov chain model corresponding to the health state of the mechanical gearbox based on a dynamic modeling method; As a preferred embodiment, S3 includes: S31, determining the state space, and dividing the health state of the gearbox into {normal, mild degradation, moderate degradation, severe failure}; In this embodiment, the health state of the gearbox is classified. For example, it can be divided into several states such as normal, mild degradation, moderate degradation, and severe failure, which are represented by S 1 、S 2 、S 3 、S 4 respectively, so as to determine the state space {S 1 ,S 2 ,S 3 ,S 4}.

[0029] S32. Obtain the collected multi-source data; S33. Divide the time interval, including: according to the operation characteristics of the mechanical gearbox and the data acquisition frequency, select an appropriate time interval Δt, such as observing and recording the state change every 1 hour or several hours.

[0030] S34. Determine the transition probability, including: for a non-uniform Markov chain, the transition probability is related to time. According to historical data, calculate the probability P i of transitioning from state S j to state S ij at different time intervals nΔt, that is: P ij (n)=P(X (k+n)Δt =S j |X kΔt =S i ); S35. Establish a non-uniform Markov chain model corresponding to the health state of the mechanical gearbox, including: represent the transition probability matrix P(n) of the non-uniform Markov chain model in matrix form, and its elements are P ij (n). Assume that the probability of the gearbox being in state S i at time kΔt is π i (k), then the probability π j of being in state S j at time (k + n)Δt can be calculated by: π j (k + n)=sum(π i (k)P ij )(where i = 1, 2, 3, 4).

[0031] S36. Verify and optimize the non-uniform Markov chain model, including: (1) Verify the established non-uniform Markov chain model using new experimental data, and evaluate the accuracy of the model by comparing the index differences between the model prediction results and the actual observation results; among them, the index differences include mean square error index differences; (2) If the non-uniform Markov chain model is inaccurate, analyze the reasons and optimize the non-uniform Markov chain model; wherein, the optimization includes: readjusting the state space and / or improving the calculation method of the transition probability.

[0032] S4. Based on the Bayesian network-fault mode mapping and the non-uniform Markov chain model, perform dynamic prediction and decision optimization of the remaining life of the mechanical gearbox; As a preferred embodiment, the S4 includes: S41. Based on the Viterbi algorithm, backtrack the optimal state path, and calculate the expected time from the current state to the failure threshold and the 95% confidence interval of the remaining life; S42. Input the real-time data stream into the Bayesian network after feature extraction, and output the probability of failure occurrence based on the Bayesian network-fault mode mapping; S43. Based on the probability of failure occurrence and the working condition parameters of the mechanical gearbox, drive the state transition of the non-uniform Markov chain model, and update the prediction result of the remaining life of the mechanical gearbox; S44. Based on the comparison between the 95% confidence interval of the remaining life and the maintenance economy model, trigger an early warning or generate a maintenance suggestion, and provide decision optimization according to the state of the mechanical gearbox.

[0033] As a preferred embodiment, the backtracking of the optimal state path in the S41 includes: (1) Initialization: Based on the known non-uniform Markov chain model, set the state probability distribution π 0 , that is, the probability that the gearbox is in each state at the initial moment; For each state i, set the initial path metric δ 0 (i) = π 0 (i); And record the predecessor state ψ of the initial state 0 (i) = 0; (2) Perform recursive calculation: Starting from the initial moment, at each time step t, for each state j, calculate the path metric δ from all possible previous states i to state j t (j) = max i [δ t-1 (i)P ij (t)]; At the same time, record the predecessor state ψ that maximizes this path metric t (j) = argmax i [δ t-1 (i)P ij (t)], where P ij(t) is the transition probability from state i to state j at time t.

[0034] (3) Determine the termination conditions, including: stop the calculation when reaching a certain termination time or reaching the state corresponding to the failure threshold. Assume that the termination condition is reached at time T, and find the state i that maximizes δ T (i) * , and this state is the final optimal state.

[0035] (4) Backtrack the path: Starting from the final optimal state i * and using the recorded predecessor state information ψ t (i), backtrack to obtain the optimal state path {i * T , i * T-1 ,…i * 0}.

[0036] As a preferred embodiment, the calculation of the expected time from the current state to the failure threshold in S41 includes: (1) Determine the transition time probability distribution: According to the non-uniform Markov chain model, determine the time probability distribution required to transfer from each state to other states. The probability density function of the state transition time (such as exponential distribution, Weibull distribution, etc.) can be fitted through historical data.

[0037] (2) Expected time calculation: Starting from the current state and along the optimal state path, calculate the expected time of each state transition according to the state transition time probability distribution. Let the expected time from state i to state j be E[t ij , along the optimal state path {i * T , i * T-1 ,…i * 0}, the expected time E[T] from the current state (assumed to be i * 0 ) to the failure threshold (assuming the final state is i * T ) is E[T]=sumE[t i*ti*t+1 (t = 0,…T).

[0038] As a preferred embodiment, the calculation of the 95% confidence interval of the remaining life in S41 includes: (1) Simulation sampling: Using the non-uniform Markov chain model and the state transition time probability distribution, conduct a large number of simulations (such as N times). Each simulation starts from the current state and along the randomly generated state path, calculate the time T to reach the failure threshold (k), where k = 1, 2, …, N; (2) Sorting and confidence interval calculation: Sort the times T obtained from N simulations (k) in ascending order; According to the definition of the confidence interval, the lower limit of the 95% confidence interval is the 0.025N-th value after sorting, and the upper limit is the 0.975N-th value (when N is large). That is, the 95% confidence interval of the remaining life is [T lower , T upper , where T lower and T upper are the calculated lower and upper limit values respectively.

[0039] As Figure 2 shown, this embodiment provides a dynamic reliability evaluation system for a large track maintenance machine gearbox to implement the method of Embodiment 1. The system includes: A data acquisition and feature extraction module 101 for multi-source data acquisition and feature extraction; A Bayesian network establishment module 102 for establishing a Bayesian network - fault mode mapping based on the extracted features; A non-uniform Markov chain model establishment module 103 for establishing a non-uniform Markov chain model corresponding to the health state of the mechanical gearbox based on a dynamic modeling method; A reliability evaluation module 104 for performing dynamic prediction of the remaining life and decision optimization of the mechanical gearbox based on the Bayesian network - fault mode mapping and the non-uniform Markov chain model.

[0040] The present invention also provides a memory storing multiple instructions for implementing the method as in Embodiment 1.

[0041] As Figure 3 shown, the present invention also provides an electronic device including a processor 301 and a memory 302 connected to the processor 301. The memory 302 stores multiple instructions that can be loaded and executed by the processor so that the processor can execute the method as in Embodiment 1.

[0042] Finally, it should be noted that: The above embodiments are only used to illustrate the technical solutions of the present invention, not to limit them; Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: They can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; And these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dynamic reliability assessment method for a large-scale road maintenance machinery gearbox, characterized in that: The following steps are involved: S1, multi-source data collection and feature extraction; S2, establishing a Bayesian network-fault mode mapping based on the extracted features; S3, establishing a non-uniform Markov chain model corresponding to the health state of the mechanical gearbox based on a dynamic modeling method; S4, performing dynamic prediction and decision optimization of the remaining life of the mechanical gearbox based on the Bayesian network-fault mode mapping and the non-uniform Markov chain model.

2. A large-scale road maintenance machinery gearbox dynamic reliability assessment method according to claim 1, characterized in that: The S1 includes: S11, performing multi-source data collection, including: deploying a vibration acceleration sensor, a temperature sensor, and an oil particle counter, and collecting the gearbox vibration time domain waveform, gearbox temperature, and lubricating oil wear particle concentration under different operating conditions based on the deployed vibration acceleration sensor, temperature sensor, and oil particle counter; and recording the actual health status change of the mechanical gearbox at the same time; S12, extracting features from multi-source data, including: performing wavelet decomposition on the gearbox vibration time domain waveform, and extracting feature values ​​as frequency domain degradation indicators.

3. A large-scale road maintenance machinery gearbox dynamic reliability assessment method according to claim 2, characterized in that: The S2 includes: S21, establishing a network topology diagram of a Bayesian network, wherein the network topology diagram of the Bayesian network includes three defined layers of nodes, each layer of nodes includes a root node and multiple leaf nodes, the root node is used to characterize a fault mode; the multiple leaf nodes are used to characterize health indicators and operating parameters; wherein the fault mode includes: F1: tooth surface pitting, F2: bearing spalling, F3: cage damage; the health indicators include: H1: vibration, H2: temperature, H3: oil particles; the operating parameters include: C1: load rate, C2: ambient temperature; S22, calculating a plurality of conditional probability tables, including: based on a historical fault case library, learning conditional probability relationships between a plurality of root nodes and the plurality of leaf nodes through a data mining algorithm, thereby forming the plurality of conditional probability tables; S23, establishing a fault-feature-operating condition causal reasoning network based on the network topology diagram and the multiple conditional probability tables, wherein the fault-feature-operating condition causal reasoning network includes the Bayesian network-fault mode mapping.

4. A method for dynamic reliability assessment of a large-scale road maintenance machinery gearbox according to claim 3, characterized in that: The S3 includes: S31, determine the state space, divide the health state of the gearbox into {normal, slightly degraded, moderately degraded, severe fault}, represented by S1, S2, S3, S4 respectively, so as to determine the state space {S1, S2, S3, S4}; S32, acquiring the collected multi-source data; S33, dividing the time interval, including: selecting a suitable time interval Δt according to the operating characteristics of the mechanical gearbox and the data collection frequency; S34, determining the transition probability, including: calculating the transition probability from state S at different time intervals nΔt according to historical data i Transfer to state S j The probability P ij (n), namely: P ij (n)=P(X (k+n)Δt =S j |X kΔt =S i ); S35, establishing a non-uniform Markov chain model corresponding to the health state of the mechanical gearbox, including: expressing the transition probability matrix P(n) of the non-uniform Markov chain model in matrix form, whose elements are P ij (n); Assume that at time kΔt, the gearbox is in state S i The probability is π i (k), then at time (k+n)Δt it is in state S j The probability π j (k+n) can be obtained by: π j (k+n)=sum(π i (k)P ij (n)) calculation, where i=1, 2, 3, 4; S36, verification and optimization of non-uniform Markov chain model.

5. A large-scale road maintenance machinery gearbox dynamic reliability assessment method according to claim 4, characterized in that: The S36 includes: (1) The established non-uniform Markov chain model is verified using new experimental data, and the accuracy of the model is evaluated by comparing the index differences between the model prediction results and the actual observation results; the index differences include the mean square error index difference' (2) If the non-uniform Markov chain model is inaccurate, analyzing the cause and optimizing the non-uniform Markov chain model; wherein the optimization includes: readjusting the state space and / or improving the calculation method of the transition probability.

6. A large-scale road maintenance machinery gearbox dynamic reliability assessment method according to claim 5, characterized in that: The S4 includes: S41, based on the Viterbi algorithm, trace back the optimal state path, calculate the expected time from the current state to the failure threshold and calculate the 95% confidence interval of the remaining life; S42, extracting features from the real-time data stream and inputting it into the Bayesian network, and outputting the probability of a fault occurring based on the Bayesian network-fault mode mapping; S43, driving the state transition of the non-uniform Markov chain model based on the fault occurrence probability and the operating parameters of the mechanical gearbox, and updating the remaining life prediction result of the mechanical gearbox; S44, triggering an early warning or generating a maintenance suggestion based on a comparison between the 95% confidence interval of the remaining life and the maintenance economy model, and providing decision optimization according to the state of the mechanical gearbox.

7. A method for dynamic reliability assessment of a large-scale road maintenance machinery gearbox according to claim 6, characterized in that: The optimal state path backtracking based on the Viterbi algorithm in S41 includes: (1) Initialization: Based on the known non-uniform Markov chain model, the state probability distribution π0 at the initial moment is set, that is, the probability of the gearbox being in each state at the initial moment; For each state i, set the initial path metric δ0(i)=π0(i); And record the predecessor state of the initial state ψ0(i)=0; (2) Perform recursive calculation: Starting from the initial moment, at each time step t, for each state j, calculate the path metric δ from all possible previous states i to state j t (j)=max i [δ t-1 (i)P ij (t)]; At the same time, record the predecessor state ψ that maximizes the path metric t (j)=argmax i [δ t-1 (i)P ij (t)], where P ij (t) is the transition probability from state i to state j at time t; (3) Determine the termination condition, including: stop the calculation when a certain termination time is reached or the state corresponding to the failure threshold is reached. Assuming that the termination condition is reached at time T, find the condition that makes δ T (i) Maximum state i * , this state is the final optimal state; (4) Backtracking path: From the final optimal state i * Start by using the recorded predecessor state information ψ t (i), backtrack to get the optimal state path {i * T ,i * T-1 ,…i * 0}.

8. A method for dynamic reliability assessment of a large-scale road maintenance machinery gearbox according to claim 7, characterized in that: The calculation of the expected time from the current state to the failure threshold in S41 includes: (1) Determine the probability distribution of transition time: Based on the non-uniform Markov chain model, determine the probability distribution of the time required to transition from each state to other states, and fit the probability density function of the state transition time through historical data; (2) Calculation of expected time: Starting from the current state, along the optimal state path, the expected time of each state transition is calculated according to the probability distribution of state transition time; let the expected time from state i to state j be E[t ij ], along the optimal state path {i * T ,i * T-1 ,…i * 0}, the current state (assuming i * 0) to the failure threshold (assuming the final state is i * T )’s expected time E[T]=sumE[t i*ti*t+1 ](t=0,…T).

9. A method for dynamic reliability assessment of a large-scale road maintenance machinery gearbox according to claim 8, characterized in that: The calculation of the 95% confidence interval of the remaining life in S41 includes: (1) Simulation sampling: Using the non-uniform Markov chain model and the state transition time probability distribution, a large number of simulations (such as N times) are performed. Each simulation starts from the current state and follows the randomly generated state path to calculate the time T to reach the failure threshold. (k) , where k=1,2,…,N; (2) Sorting and confidence interval calculation: The time T obtained by N simulations (k) Sort from small to large; according to the definition of confidence interval, the lower limit of the 95% confidence interval is the 0.025Nth value after sorting, and the upper limit is the 0.975Nth value; that is, the 95% confidence interval of the remaining life is [T lower , T upper ], where T lower and T upper are the calculated lower and upper limits, respectively.

10. A dynamic reliability assessment system for a large-scale road maintenance machinery gearbox, used to implement the method described in any one of claims 1 to 9, characterized in that: The system comprises: A data collection and feature extraction module (101), used for multi-source data collection and feature extraction; A Bayesian network building module (102), used to build a Bayesian network-fault mode mapping based on the extracted features; A non-uniform Markov chain model building module (103), used to build a non-uniform Markov chain model corresponding to the health state of the mechanical gearbox based on a dynamic modeling method; A reliability assessment module (104) is used to dynamically predict the remaining life of the mechanical gearbox and optimize the decision based on the Bayesian network-fault mode mapping and the non-uniform Markov chain model.

Images