Mechanical system fault causal network modeling method based on mechanism model
The parameterized causal directed graph is constructed through the mechanism model and converted into actual failure events, which solves the shortcomings of causal modeling of mechanical system failures with lack of experience and data, and achieves a causal network model with high integrity and accuracy.
Patent Information
- Application Number
- CN202510561297.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-30
- Publication Date
- 2025-08-12
AI Technical Summary
The existing technology has insufficient applicability in the causal relationship modeling of mechanical system failures with lack of experience and data, and it is difficult to effectively build a complete and accurate causal network model.
The mechanism model is used to establish a fault mechanism model, a parameterized causal directed graph is constructed through mathematical structure transformation, and a fishbone graph analysis method is used to convert physical parameters into actual fault events to generate a causal network model.
Reliance on experience and data is reduced, the integrity and accuracy of the fault causal network model is improved, and it is suitable for mechanical system fault traceability scenarios where data is scarce or inexperienced.
Smart Images
Figure CN120470909A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of mechanical system failure causal modeling, and in particular to a mechanical system failure causal network modeling method based on a mechanism model. Background Art
[0002] Causal modeling of mechanical system failures is a fundamental component of fault tracing and diagnosis. The quality of its construction directly impacts the accuracy of root cause location and diagnostic efficiency. In equipment operation and maintenance, when a mechanical system experiences an abnormal operating condition, technicians must first construct a causal network model that characterizes the relationship between the failure mode and potential causes. They then use reasoning methods to reversely trace the fault propagation path. Therefore, the integrity and accuracy of the causal model directly determine the engineering value of the fault diagnosis system.
[0003] Current mainstream fault causal modeling methods can be categorized as experience-driven and data-driven. Traditional methods often employ expert-based modeling paradigms, such as failure mode and effects analysis (FMEA) and fault tree analysis (FTA). While these methods offer the advantage of operational convenience, their modeling quality is highly dependent on the analyst's prior knowledge and accumulated domain experience, resulting in inherent flaws such as strong subjectivity and poor reproducibility. With the development of intelligent operation and maintenance technologies, data-driven methods based on machine learning have gained widespread application in fault diagnosis. While these methods reduce reliance on experience, they face the following technical bottlenecks: First, the modeling process requires massive amounts of high-quality training data, placing stringent requirements on sensor deployment density, data collection continuity, and feature engineering completeness; second, under complex operating conditions, the modeling process is susceptible to interference from environmental noise, leading to misjudgments of causal relationships; and finally, data-driven models exhibit a "black box" nature, which limits the interpretability of diagnostic results.
[0004] It is particularly important to note that existing technical solutions face significant application scenario limitations during implementation: experience-driven methods struggle to adapt to the fault modeling needs of new and complex equipment, while data-driven methods also face implementation barriers in specialized scenarios such as low-speed, heavy-load equipment and small-batch customized equipment, where data acquisition costs are high. This dual technical limitation significantly reduces the applicability of traditional methods in industrial scenarios where empirical knowledge reserves are insufficient or data collection conditions are limited. Summary of the Invention
[0005] In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is: how to provide a modeling method for the causal relationship of mechanical system failures that is suitable for lack of experience and data, which can reduce the dependence of the modeling process on experience and data and provide a basis for tracing and diagnosing mechanical product failures.
[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:
[0007] A mechanical system fault causal network modeling method, characterized by comprising the following steps:
[0008] Step S1: Analyze the working principle of the target mechanical system and establish a fault mechanism model based on the mechanism analysis method for the fault mode to be analyzed;
[0009] Step S2: performing mathematical structure transformation on the fault mechanism model to construct a parameterized causal directed graph, including:
[0010] S2.1) performing mathematical transformation on the fault mechanism model to construct a complete structure in the form of algebraic equations or first-order differential equations, with the total number of equations equal to the total number of system variables;
[0011] S2.2) Recursively divide the complete structure into k-order minimum complete structure subsets by a multi-order decomposition method k E s and the remaining subset k E i , and define endogenous variables and exogenous variables of each order based on the intersection relationship of variables;
[0012] The minimum complete structure subset is an equation subset with the same total number of equations and variables and the least number of variables;
[0013] S2.3) constructing a symbolically labeled qualitative causal relationship diagram based on the parameter causal criteria;
[0014] For the endogenous and exogenous variables of each order, the following parameter causal criterion is adopted. With the first-order exogenous variable as the starting point and the highest-order endogenous variable (as the characteristic parameter of the fault mode) as the end point, one-way arrows are used to point from lower-order variables to higher-order variables step by step to obtain a parameter causal relationship diagram. If the variable exists in a differential form in the complete structure, it is converted into a special solution of the variable with respect to time under a certain initial state, and the differential initial variable and action time parameter are introduced. Based on the fault mechanism model, the positive and negative correlations between adjacent variables in the parameter causal relationship diagram are judged and symbolized respectively. Parameters whose correlation cannot be determined are differentially marked, and finally a parameterized causal directed graph for the fault mode is generated.
[0015] The parameter causality criterion is: if and only if the exogenous variable in the k-order complete structure k Ex and endogenous variables k When the union of En constitutes all variables of this order, k Each variable in Ex is k The direct cause of the corresponding variable in En;
[0016] Step S3: converting the physical parameters in the parameterized causal directed graph into actual fault events to construct a causal network model;
[0017] Based on the parameterized causal directed graph, the qualitative influence relationships between first-order exogenous variables, differential initial variables, and action time on the highest-order endogenous variables are established and symbolically labeled. A fishbone diagram analysis method is used to convert the physical parameters in the directed graph into actual fault events. The fault mode and its parameterized representation are placed in the fish head, while the first-order exogenous variables, differential initial variables, action time, and their qualitative influence relationships on the fault mode parameters are placed in the fish bones. The fishbone portion on each fish bone represents the potential root cause of the fault mode. Correlation analysis is used to determine the events corresponding to each fishbone portion. These events cause each first-order exogenous variable, differential initial variable, and action time to trigger changes in the fault mode parameters. Ultimately, a causal network model for the fault mode is obtained.
[0018] Preferably, the mechanism analysis method includes kinematic analysis and dynamic analysis methods, specifically characterizing the fault evolution process by establishing multi-body system motion equations, Lagrange equations and energy conservation equations.
[0019] Preferably, the mathematical transformation processing includes using an order reduction method to eliminate high-order differential terms, introducing intermediate state variables, analytical solution or introducing a constant term.
[0020] Preferably, the variable set of the complete structure includes:
[0021] (a) Part intrinsic parameters: geometric dimensions, material properties, center of mass, mass, moment of inertia, and speed ratio.
[0022] (b) Load parameters: driving force, driving torque, resistance, and resistance torque.
[0023] (c) State parameters: speed, rotation speed, displacement, and rotation angle.
[0024] Among them, the time parameter in the differential equation is not included in the total as an independent variable.
[0025] Preferably, the multi-order decomposition method is specifically:
[0026] The initial complete structure is denoted as 0 E, for the kth order complete structure k E, execute:
[0027] (1) k E is divided into the minimum complete structure subset k E s and the remaining subset k E i ;
[0028] (2)k E s and k E i Substitute the common variables into k E i Generate a (k+1)-order complete structure;
[0029] (3) Recursively execute until the remaining subset is empty.
[0030] Among them, a variable is determined to be an endogenous variable if and only if it belongs to the kth order complete structure and does not belong to all lower-order minimum subsets, and an exogenous variable if and only if it belongs to the kth order complete structure and belongs to any lower-order minimum subset.
[0031] Compared with the prior art, the present invention has the following beneficial effects:
[0032] This paper proposes a novel causal network modeling method for mechanical system failures based on a mechanism model. This method establishes a parameterized causal relationship between failure modes and root causes based on the mechanism model. These abstract physical parameters are then mapped to actual failure events to create a causal network model of the mechanical system failure. This method does not rely heavily on field data or empirical knowledge and is applicable to data-scarce or inexperienced mechanical system failure tracing scenarios. By guiding analysts through targeted causal modeling through changes in physical parameters, it effectively reduces the risk of missing potential failure causes and improves the integrity of the causal network model. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Figure 1 This is a flow chart of the fault causal network modeling of the present invention.
[0034] Figure 2 Schematic diagram of the fishbone diagram in the embodiment.
[0035] Figure 3 Schematic diagram of qualitative causal relationship of parameters in the embodiment.
[0036] Figure 4 This is the causal network model of the “out-of-position rotation” failure of the pallet exchange rack in the embodiment. DETAILED DESCRIPTION
[0037] The present invention will be described in further detail below with reference to the accompanying drawings.
[0038] A mechanism-driven causal network modeling method for mechanical system failures, the process is as follows Figure 1 As shown in the figure, first, a fault mechanism model is established for the fault mode to be analyzed in the mechanical system; secondly, a causal directed graph of the physical parameters of the mechanical system is established based on the mechanism model; finally, the fishbone diagram method is used to convert the parameter causal directed graph into a set of specific fault events to form a fault causal network model.
[0039] The detailed steps of this method are as follows:
[0040] Step S1: Analyze the working principle of the target mechanical system. For the fault mode to be analyzed, establish the multi-body system motion equation, Lagrangian equation or energy conservation equation to characterize its fault evolution process based on kinematic analysis and dynamic analysis methods.
[0041] Step S2: performing mathematical structure transformation on the fault mechanism model to construct a parameterized causal directed graph, including:
[0042] S2.1) Processing the fault mechanism model using order reduction, introduction of intermediate variables, analytical solution, or introduction of constant terms to construct a complete structure composed of algebraic equations or first-order differential equations with a total number of equations equal to the total number of system variables;
[0043] Preferably, the variable set of the complete structure includes:
[0044] (a) Part intrinsic parameters: geometric dimensions, material properties, center of mass, mass, moment of inertia, and speed ratio.
[0045] (b) Load parameters: driving force, driving torque, resistance, and resistance torque.
[0046] (c) State parameters: speed, rotation speed, displacement, and rotation angle.
[0047] Among them, the time parameter in the differential equation is not included in the total as an independent variable.
[0048] S2.2) Recursively divide the complete structure into k-order minimum complete structure subsets by a multi-order decomposition method k E s and the remaining subset k E i , and define endogenous variables and exogenous variables of each order based on the intersection relationship of variables;
[0049] The multi-order decomposition method is specifically:
[0050] The initial complete structure is denoted as 0 E, for the kth order complete structure k E, execute:
[0051] (1) k E is divided into the minimum complete structure subset k E s and the remaining subset k E i ;
[0052] (2) k E s andk E i Substitute the common variables into k E i Generate a (k+1)-order complete structure;
[0053] (3) Recursively execute until the remaining subset is empty.
[0054] Among them, a variable is determined to be an endogenous variable if and only if it belongs to the kth order complete structure and does not belong to all lower-order minimum subsets, and an exogenous variable if and only if it belongs to the kth order complete structure and belongs to any lower-order minimum subset;
[0055] The minimum complete structure subset is an equation subset in which the total number of equations is equal to the total number of variables and has the least variables.
[0056] S2.3) constructing a symbolically labeled qualitative causal relationship diagram based on the parameter causal criteria;
[0057] For the endogenous and exogenous variables of each order, the following parameter causal criterion is adopted. With the first-order exogenous variable as the starting point and the highest-order endogenous variable (as the characteristic parameter of the fault mode) as the end point, one-way arrows are used to point from lower-order variables to higher-order variables step by step to obtain a parameter causal relationship diagram; if the variable exists in a differential form in the complete structure, it is converted into a special solution of the variable with respect to time under a certain initial state, and the differential initial variable and the action time parameter are introduced; based on the fault mechanism model, the positive and negative correlations between adjacent variables in the parameter causal relationship diagram are judged, and symbolic markings including positive and negative signs, color coding, linear transformation, digital identification and arrow annotation are respectively performed on them; parameters whose correlation cannot be determined are differentially marked, and finally a parameterized causal directed graph for the fault mode is generated;
[0058] The parameter causality criterion is: if and only if the exogenous variable in the k-order complete structure k Ex and endogenous variables k When the union of En constitutes all variables of this order, k Each variable in Ex is k The direct cause of the corresponding variable in En;
[0059] Step S3: converting the physical parameters in the parameterized causal directed graph into actual fault events to construct a causal network model;
[0060] Based on the parameterized causal directed graph, the qualitative influence relationship of the first-order exogenous variable, the differential initial variable and the action time on the highest-order endogenous variable is established and symbolized; Figure 2The fishbone diagram analysis method shown here converts the physical parameters in the directed graph into actual fault events. The fault mode and its parameterized representation are placed in the fish head, while the first-order exogenous variables, differential initial variables, and action time, along with their qualitative influence on the fault mode parameters, are placed in the fish bones. The fishbone on each fishbone represents the potential root cause of the fault mode. Correlation analysis identifies the events corresponding to each fishbone, which cause the first-order exogenous variables, differential initial variables, and action time to trigger changes in the fault mode parameters. Ultimately, a causal network model for the fault mode is derived.
[0061] The present invention will be further described in detail below with reference to specific embodiments.
[0062] according to Figure 1 The process shown in the figure establishes a fault causal network for the failure mode of the CNC machine tool pallet exchange rack "not rotating in place", which includes the following steps:
[0063] (a) The structure and working principle of the CNC machine tool pallet exchange rack are analyzed, and the rotational dynamics equations are established using the D'Alembert principle as shown in formulas (1) to (3):
[0064]
[0065] in, is the corner position of the pallet exchange rack; J e and M e are the equivalent moment of inertia and equivalent torque of the pallet exchange rack respectively; M i is the resultant moment acting on part i; i J is the speed ratio between the pallet and part i on the motion transmission path; Si is the moment of inertia of part i about its center of mass, and K is the number of movable rigid parts that make up the pallet exchange rack structure.
[0066] (b) Since there are only three equations in formula (1)-formula (3) but six variables, formula (1)-formula (3) are not complete structures. By introducing intermediate variables, formula (1) is converted into two first-order differential equations, and formula (4) and formula (5) are obtained. In addition, M i and J Si It can be expressed using basic parameters to obtain formula (6) and formula (7).
[0067]
[0068] Where, ω is the rotation speed of the pallet exchange rack; and are the driving torque and resistance torque acting on part i respectively; m i is the mass of part i; r iis the perpendicular distance between the center of mass of part i and the axis of rotation.
[0069] Since the parameter ξ i ,r i ,m i , and It is determined according to the physical structure and working conditions of the pallet exchange rack and can be treated as a constant.
[0070] ξ i =c ξi , (8)
[0071] r i =c ri , (9)
[0072] m i =c mi , (10)
[0073]
[0074] So far, there are exactly 11 equations and 11 variables in formula (2) to formula (12), forming a complete structure. Among them, the intrinsic parameters of the part include m i , r i ,ξ i , J Si , J e ; Load parameters include M i , M e ; The state parameters include ω, The action time t is not included in the total number of variables.
[0075] (c) By decomposing the above complete structure step by step, we can obtain the complete structure of each order of the CNC machine tool pallet exchange rack and its endogenous and exogenous variables, as shown in Table 1.
[0076] Table 1 The complete structure of each order of pallet exchange rack and its endogenous and exogenous variables
[0077]
[0078] (d) For the differential forms of formula (4) and formula (5), add the differential initial variables ω0 and Based on formula (2)-formula (12), the qualitative correlation between the parameters can be obtained, and the solid line and dotted line respectively represent the positive correlation and negative correlation between two adjacent variables, and the following is obtained: Figure 3 Qualitative causal relationship diagram of each parameter is shown.
[0079] (e) Establish first-order exogenous variables based on the parameter qualitative causal relationship diagram Differentiation initial variable and action time (t) on the rotation angle of the pallet exchange rack The upward arrow "↑" indicates that a larger parameter will lead to a failure mode at the fish head (i.e., the pallet exchange rack "does not rotate in place"), and the downward arrow "↓" indicates that a smaller parameter will lead to a failure mode at the fish head. The causal network model of the pallet exchange rack "does not rotate in place" is established through the fishbone diagram, as shown in the following example: Figure 4 shown.
[0080] The present invention's causal network modeling method for mechanical system failures, based on a mechanism model, utilizes a parameter causal method to establish a directed causal graph between underlying physical parameters related to mechanical system failures, obtaining the qualitative influence relationships of the most original underlying physical parameters that cause the failure mode to occur. The method then converts the qualitative changes in the underlying parameters into specific events related to the mechanical system structure and environmental conditions, thereby obtaining a causal network model for the failure mode. This method guides analysts in conducting targeted causal modeling through changes in physical parameters, effectively reducing the risk of missing potential causes of failures. Furthermore, because this method establishes causal relationships through physical parameters, it does not require excessive reliance on experience and data and is therefore applicable to causal network modeling of mechanical system failures where historical data is scarce or inexperienced.
[0081] Finally, it should be noted that the above-described embodiments of the present invention are merely examples for illustrating the present invention and are not intended to limit the embodiments of the present invention. Although the applicant has described the present invention in detail with reference to preferred embodiments, a person skilled in the art will be able to make other variations and modifications based on the above description. It is not possible to enumerate all embodiments here. Any obvious changes or modifications derived from the technical solution of the present invention remain within the scope of protection of the present invention.
Claims
1. A causal network modeling method for mechanical system failure based on a mechanism model, characterized in that: The following steps are involved: Step S1: Analyze the working principle of the target mechanical system and establish a fault mechanism model based on the mechanism analysis method for the fault mode to be analyzed; Step S2: performing mathematical structure transformation on the fault mechanism model to construct a parameterized causal directed graph, including: S2.1) performing mathematical transformation on the fault mechanism model to construct a complete structure in the form of algebraic equations or first-order differential equations, with the total number of equations equal to the total number of system variables; S2.2) Recursively divide the complete structure into k-order minimum complete structure subsets by a multi-order decomposition method k E s and the remaining subset k E i , and define endogenous variables and exogenous variables of each order based on the intersection relationship of variables; The minimum complete structure subset is an equation subset with the same total number of equations and variables and the least number of variables; S2.3) constructing a symbolically labeled qualitative causal relationship diagram based on the parameter causal criteria; For the endogenous and exogenous variables of each order, the following parameter causal criterion is adopted. With the first-order exogenous variable as the starting point and the highest-order endogenous variable (as the characteristic parameter of the fault mode) as the end point, one-way arrows are used to point from lower-order variables to higher-order variables step by step to obtain a parameter causal relationship diagram. If the variable exists in a differential form in the complete structure, it is converted into a special solution of the variable with respect to time under a certain initial state, and the differential initial variable and action time parameter are introduced. Based on the fault mechanism model, the positive and negative correlations between adjacent variables in the parameter causal relationship diagram are judged and symbolized respectively. Parameters whose correlation cannot be determined are differentially marked, and finally a parameterized causal directed graph for the fault mode is generated. The parameter causality criterion is: if and only if the exogenous variable in the k-order complete structure k Ex and endogenous variables k When the union of En constitutes all variables of this order, k Each variable in Ex is k The direct cause of the corresponding variable in En; Step S3: converting the physical parameters in the parameterized causal directed graph into actual fault events to construct a causal network model; Based on the parameterized causal directed graph, the qualitative influence relationships between first-order exogenous variables, differential initial variables, and action time on the highest-order endogenous variables are established and symbolically labeled. A fishbone diagram analysis method is used to convert the physical parameters in the directed graph into actual fault events. The fault mode and its parameterized representation are placed in the fish head, while the first-order exogenous variables, differential initial variables, action time, and their qualitative influence relationships on the fault mode parameters are placed in the fish bones. The fishbone portion on each fish bone represents the potential root cause of the fault mode. Correlation analysis is used to determine the events corresponding to each fishbone portion. These events cause each first-order exogenous variable, differential initial variable, and action time to trigger changes in the fault mode parameters. Ultimately, a causal network model for the fault mode is obtained.
2. The method according to claim 1, wherein: The mechanism analysis method includes kinematic analysis and dynamic analysis methods, and specifically characterizes the fault evolution process by establishing the multi-body system motion equation, Lagrange equation and energy conservation equation.
3. The method according to claim 1, wherein: The mathematical transformation processing includes adopting the reduction method to eliminate high-order differential terms, introducing intermediate state variables, analytical solution and introducing constant terms.
4. The method according to claim 1, wherein: The variable set of the complete structure includes part intrinsic parameters, load parameters and state parameters, and the time parameter in the differential equation is not included in the total as an independent variable.
5. The method according to claim 1, wherein: The multi-order decomposition method is specifically: The initial complete structure is denoted as 0 E, for the kth order complete structure k E, execute: (1) k E is divided into the minimum complete structure subset k E s and the remaining subset k E i ; (2) k E s and k E i Substitute the common variables into k E i Generate a (k+1)-order complete structure; (3) Recursively execute until the remaining subset is empty. Among them, a variable is determined to be an endogenous variable if and only if it belongs to the kth order complete structure and does not belong to all lower-order minimum subsets, and an exogenous variable if and only if it belongs to the kth order complete structure and belongs to any lower-order minimum subset.