A condition monitoring and fault diagnosis method based on digital twin perception framework

Abstract

The present invention relates to a condition monitoring and fault diagnosis method based on a digital twin perception framework, comprising an offline phase and an online phase. The offline phase comprises: dividing a numerical calculation discrete model according to the physical structure of the industrial equipment to be tested; determining the partial differential equation of the corresponding physical field based on the numerical calculation discrete model, and collecting and acquiring a data set of the physical field; performing data analysis on the data set to realize real-time calculation of the high-dimensional physical field, and then constructing a reliability assessment system, a physical filtering system, and an abnormal working condition identification system respectively to form a digital twin perception framework; the online phase comprises the following steps: obtaining the measured data of the industrial equipment to be tested, inputting it into the digital twin perception framework, and performing reliability assessment, physical filtering, and abnormal working condition identification. Compared with the prior art, the present invention has high computational efficiency and is interpretable, and can efficiently and accurately perform condition monitoring and fault diagnosis.

Description

Technical Field

[0001] The present invention relates to the field of equipment status monitoring, and in particular to a status monitoring and fault diagnosis method based on a digital twin perception framework. Background Art

[0002] The main goals of equipment operation and maintenance are to ensure long-term, stable, and efficient performance while being able to respond quickly when faults occur. Key technologies in this area include condition monitoring and fault diagnosis, which have been widely used in industrial environments, especially in areas where equipment reliability and life are critical, such as petrochemicals, manufacturing, and power systems. In particular, for components operating under extreme conditions, the number of sensors is strictly limited. For example, in high-temperature rotors, the installation of sensors is extremely difficult, resulting in extremely sparse data. This makes it extremely challenging to obtain reliable, physically interpretable high-dimensional physical information and effective complex fault diagnosis in real time. However, in industrial environments, components operating under extreme conditions are usually high-cost and high-risk. It is these components that have the most urgent need for efficient and accurate condition monitoring and fault diagnosis.

[0003] Currently, there are three main types of rapid assessment technologies for condition monitoring and fault diagnosis:

[0004] The first is traditional condition monitoring and fault diagnosis technology. This technology uses sensors and data acquisition systems to collect key equipment signals (such as vibration, temperature, pressure, and noise) in real time. Notably, due to the limited number of sensors, these signals are low-dimensional. Data-driven algorithms or empirical methods are then used to analyze and process the collected signals, generating reference low-dimensional data that can reveal the health status of the equipment. However, using low-dimensional reference data as a basis for condition monitoring and fault diagnosis is often unreliable.

[0005] The second approach is to implement real-time condition monitoring using data-driven reduced-order models. Data-driven reduced-order models use numerical simulation results as a dataset and perform proper orthogonal decomposition (POD) on the dataset to obtain a reduced-order orthogonal basis Φ for the physical field. There are two technical approaches to inverting high-dimensional physical fields from low-dimensional sensor signals. The first uses machine learning, Gaussian regression, or related fitting techniques to establish a nonlinear relationship between low-dimensional sensor signals and low-dimensional orthogonal basis coefficients, thereby achieving a mapping between the two. Then, based on the theory of proper orthogonal decomposition, the low-dimensional sensor signals are used to invert the high-dimensional physical field. The second approach uses the low-dimensional sensor signals as constraints and uses the least squares method to find the best approximation of the physical field in the reduced-order orthogonal basis space. Although both approaches are driven by numerical solutions based on physical knowledge, the solution process is still purely data-driven and lacks reliability. Furthermore, these approaches can become unstable or uncomputable when constrained by extremely sparse sensor data.

[0006] The third approach is to use a physics-driven reduced-order model for real-time computation. This method embeds a low-dimensional orthogonal basis into the physical equations to achieve dimensionality reduction. While this approach offers excellent physical interpretability and stability, it is essentially a forward computational problem requiring clear boundary conditions. To inversely solve the problem of sparse sensor data constraints, a forward iteration approach can be employed to minimize the error between numerical simulation and sensor measurement and optimize the boundary conditions. However, this significantly increases the computational effort. Especially for transient nonlinear problems, iteration is required at each time step, increasing the time cost.

[0007] Because abnormal operating conditions are often outside the data set, the results of online status assessments of abnormal operating conditions may be biased or even completely off to a certain extent. This requires reliability assessment of the calculation results, which the above three methods cannot achieve. Summary of the Invention

[0008] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a condition monitoring and fault diagnosis method based on a digital twin perception framework, using a non-invasive physical-driven inverse hyper-reduced-order model (NIPI-HROM) constrained by extremely sparse sensors, and establishing a digital twin perception framework consisting of a reliability assessment system, a physical filtering system, and an abnormal operating condition identification system with NIPI-HROM as the core.

[0009] The purpose of the present invention can be achieved by the following technical solutions:

[0010] A condition monitoring and fault diagnosis method based on a digital twin perception framework includes an offline phase and an online phase. The offline phase includes the following steps:

[0011] Divide the numerical calculation discrete model according to the physical structure of the industrial equipment to be tested;

[0012] Based on the numerical calculation discrete model, determining the partial differential equation corresponding to the physical field, and collecting and obtaining a data set of the physical field;

[0013] Performing eigendecomposition on the data set to obtain corresponding singular values ​​and eigenvectors, thereby calculating a reduced orthogonal basis of the data set;

[0014] Selecting key physical units using discrete empirical interpolation analysis based on the reduced orthogonal basis to obtain a key physical unit index matrix and a key physical unit index matrix;

[0015] The numerical calculation discrete model, reduced orthogonal basis, key physical unit index matrix and key physical unit index matrix are implanted into the partial differential equation of the physical field to realize real-time calculation of high-dimensional physical field, and then a reliability assessment system, a physical filtering system and an abnormal working condition identification system are respectively constructed to form a digital twin perception framework;

[0016] The online phase includes the following steps:

[0017] The measured data of the industrial equipment to be tested is obtained and input into the digital twin perception framework. The measured data is corrected in real time by the physical filtering system to obtain the measurement data after noise elimination. The error range of the physical field is calculated and evaluated by the reliability evaluation system. The abnormal working condition identification system is used to identify whether an abnormal working condition occurs.

[0018] Furthermore, the partial differential equation of the physical field is expressed as:

[0019]

[0020] Where R t is the nonlinear residual of the equation at time t, u is the high-dimensional physical field, f is the nonlinear differential operator, g is the nonlinear source term, x is the sensor data value, and μ is the system parameter.

[0021] Furthermore, the partial differential equation of the physical field is iteratively solved using the Newton-Raphson iterative format until the nonlinear residual approaches 0, thereby obtaining the physical field at the current moment.

[0022] Furthermore, in the iterative solution process, the calculation expression at the k-th iteration step is:

[0023]

[0024]

[0025] Where Δu is the increment of the physical field, and are the Jacobian matrix and nonlinear residual vector at the k-th iteration step at time t,

[0026] Furthermore, the calculation expression for feature decomposition of the data set is:

[0027] M=U∑V

[0028] Where M is the correlation matrix of the physical field data set at different times and under different parameterization conditions, and the columns of U correspond to MM T The eigenvector of V corresponds to M T The eigenvectors of M, ∑ are the singular values ​​s1,...,s n A diagonal matrix.

[0029] Furthermore, the calculation expression of the reduced orthogonal basis is:

[0030]

[0031]

[0032] Where Φ is a reduced orthogonal basis, v∈V=[v1,...,v r ], r is the degree of freedom of the reduced orthogonal basis, and ∈ is the weight coefficient.

[0033] Furthermore, the calculation expression of the key physical unit index matrix is:

[0034]

[0035]

[0036] Where Z p is the key physical unit index matrix, is the identity matrix No. ξ n column vectors, where

[0037] Furthermore, the numerical calculation discrete model, the reduced orthogonal basis, the key physical unit index matrix and the key physical unit index matrix are implanted into the partial differential equation of the physical field to obtain the calculation form:

[0038]

[0039]

[0040] In the formula, the sharp corner symbol represents a low-dimensional variable, is the modal coefficient increment, and is the incomplete Jacobian matrix and nonlinear residual vector of the physical key unit set and sensor constraints, w sensor is the numerical weight of the sensor, Φ is the reduced orthogonal basis, Z p is the key physical unit index matrix, Z s is the sensor unit index matrix, and u is the high-dimensional physical field.

[0041] Furthermore, the reliability evaluation system uses the sum of the nonlinear residuals of the converged physical units obtained after performing real-time calculation of the high-dimensional physical field as the physical residual R p , taking the relative L2 norm error of the physical field as E L2 , thus based on the physical residual R p and the relative L2 norm error E L2 Draw a reliability characteristic curve to evaluate the error of the calculation results.

[0042] Furthermore, the calculation expression of the reliability characteristic curve is:

[0043]

[0044] The physical filtering system determines the stability of the sensor through the reliability evaluation system and gives greater weight to stable sensor data. sensor , giving smaller weight w to unstable sensor data sensor ;

[0045] The abnormal working condition identification system is based on the physical residual R calculated by the reliability evaluation system. p The rate of change of the absolute value of the signal is compared with the preset stable rate of change range to determine whether there is a drastic change, thereby determining whether an abnormal situation occurs.

[0046] Furthermore, the industrial equipment to be tested is a high-temperature rotor to be temperature monitored.

[0047] Compared with the prior art, the present invention has the following advantages:

[0048] (1) Physical field reverse calculation: Traditional methods require the iterative method of forward calculation to achieve reverse calculation. The present invention determines the accurate boundary conditions through Newton–Raphson iteration in the offline stage. In the online stage, the NIPI-HROM of the present invention can directly perform reverse calculation without the need to calculate the accurate boundary through iteration.

[0049] (2) Reliability evaluation system: The traditional method requires a few sensor data as the basis for calculating reliability evaluation, which wastes sensor information. The present invention uses physical residuals as the basis for reliability evaluation and does not require faithful sensor information.

[0050] (3) Physical filtering system: Traditional sensor filtering algorithms are often based on statistical principles. Filtering algorithms based on different statistical preferences produce different results, resulting in weak interpretability, reliability, and physical significance. This invention uses a reliability assessment system driven by physical knowledge as the judgment standard, achieving system-level filtering functions and making substantial improvements in interpretability and filtering objects.

[0051] (4) Abnormal working condition identification system: Traditional methods use vibration modal identification to analyze the current working status of the equipment. However, the vibration signal is single and cannot obtain the physical field changes inside the equipment, making it difficult to locate the specific fault. This paper uses the reliability assessment system as the judgment standard for the abnormal working condition alarm system. It can not only provide abnormal working condition reminders but also output abnormal working condition physical fields with certain reference value, which facilitates staff to make timely and correct rescue plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is an important component of a digital twin perception framework provided in an embodiment of the present invention;

[0053] Figure 2 This is an overall architecture diagram of an online transient thermal state assessment based on a finite element reduced-order model provided in an embodiment of the present invention;

[0054] Figure 3 This is a schematic diagram of the architecture of a condition monitoring and fault diagnosis method based on a digital twin perception framework provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0056] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort shall fall within the scope of protection of the present invention.

[0057] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not require further definition or explanation in subsequent drawings.

[0058] Example 1

[0059] like Figure 3 As shown, this embodiment provides a condition monitoring and fault diagnosis method based on a digital twin perception framework, including an offline phase and an online phase. The offline phase includes the following steps:

[0060] S101: Divide the numerical calculation discrete model according to the physical structure of the industrial equipment to be tested;

[0061] Specifically: divide the numerical calculation discrete model based on the actual physical structure of industrial equipment, define unit properties and material properties, and set different types of boundary conditions according to actual working conditions.

[0062] S102: Determine a partial differential equation corresponding to a physical field based on the numerical calculation discrete model, and collect a data set of the physical field;

[0063] Specifically, we determine the partial differential equations for the relevant physical problems, establish and set parameterized sample operating conditions, and perform transient nonlinear numerical analysis. During the transient analysis, we collect data sets of the physical field at several moments - [u(t)]. The basic form of the relevant partial differential equation is expressed as:

[0064]

[0065] In the formula, u represents the desired high-dimensional physical field, f and g represent the above nonlinear differential operator and nonlinear source term, respectively. For the convenience of expression, equation (1) can be written as:

[0066]

[0067] Where R t The present invention adopts the Newton-Raphson iterative format to solve equation (2), wherein the kth iteration step is:

[0068]

[0069] Where Δu is the increment of the physical field, and are the Jacobian matrix and nonlinear residual vector at the kth iteration step at time t. Continuous iterative calculation until When it approaches 0, the high-fidelity physical field at that moment is obtained and stored in the data set.

[0070] S103: performing eigendecomposition on the data set to obtain corresponding singular values ​​and eigenvectors, thereby calculating a reduced orthogonal basis of the data set;

[0071] Singular Value Decomposition (SVD) is applied to the data set collected in step S102 to perform data compression and feature decomposition, and obtain its singular values ​​and eigenvectors, which are recorded as:

[0072] M=U∑V (4)

[0073] Where M is the correlation matrix of the data set in step S102 at different times and under different parameterization conditions, and the columns of U are the corresponding MM T The columns of V correspond to the eigenvectors of M T The eigenvectors of M, ∑ are the singular values ​​(s1, ..., s n ) is a diagonal matrix.

[0074] Obtaining a reduced orthogonal basis of the data set: The reduced orthogonal basis of the data set can be represented by the feature vectors in step S103 and the data set in step S102:

[0075]

[0076] Where v∈V=[v1,...,v r ], the degree of freedom r of the reduced-order basis is usually dozens or a few, which is much smaller than the discrete degree of freedom N (r<<N), and the weight coefficient ∈ is usually greater than 95%. The obtained reduced-order orthogonal basis Φ is then used to realize the low-dimensional linear expression of high-dimensional physical variables in orthogonal space:

[0077]

[0078] S104: Select key physical units using Discrete Empirical Interpolation Methods (DEIM) based on the reduced orthogonal basis to obtain a key physical unit index matrix and a key physical unit index matrix;

[0079] Key physical unit index matrix [Z p The calculation expression of ] is:

[0080]

[0081] in is the identity matrix No. ξ n column vector, denoted as:

[0082]

[0083] S105: implanting the numerical calculation discrete model, the reduced orthogonal basis, the key physical unit index matrix and the key physical unit index matrix into the partial differential equation of the physical field to realize real-time calculation of the high-dimensional physical field;

[0084] Specifically, based on the numerical calculation discrete model obtained in step S101, the reduced-order orthogonal basis Φ and the key physical unit index Z p and the extremely sparse sensor unit index Z s Inserted into equation (2), we get the following form:

[0085]

[0086] Among them, the angled symbols represent low-dimensional variables, is the modal coefficient increment, and is the incomplete Jacobian matrix and nonlinear residual vector of the physical key unit set and sensor constraints, w sensor is the numerical weight of the sensor. Combining equations (6) and (9) enables real-time calculation of high-dimensional physical fields.

[0087] S106: Build a reliability assessment system, a physical filtering system, and an abnormal operating condition identification system to form a digital twin perception framework.

[0088] Specifically include:

[0089] Establish a reliability evaluation system: define the sum of the nonlinear residuals of the physical unit after the convergence of equation (9) as the physical residual R p , the relative L2 norm error of the physical field is E L2 . Based on a large number of calculations, it is found that |R p | and E L2 It satisfies the linear positive correlation, and this positive correlation is named the reliability characteristic curve:

[0090]

[0091] This allows for error assessment of calculation results without relying on new sensor data. Furthermore, it is essentially driven by physical knowledge and is physically interpretable.

[0092] Establishing a physical filtering system: Different from the traditional statistical filtering algorithm, the filtering system of the present invention is established under the reliability evaluation system of physical drive in step S106. sensor The basic idea is to use the reliability evaluation system to judge the stability of the sensor and give greater weight to stable sensor data, and vice versa. sensorThe calculation process of is essentially an optimization problem, which can be calculated using various optimization calculation methods.

[0093] Establish an abnormal operating condition identification system: When | R p Drastic changes indicate an abnormal operating condition. The reliability assessment system then evaluates the error range of the physical field output by NIPI-HROM in real time, providing a crucial basis for locating the location and cause of component failure.

[0094] The online phase includes the following steps:

[0095] S201: Obtain the measured data of the industrial equipment to be tested, input it into the digital twin perception framework, and correct the measured data in real time through the physical filtering system to obtain the measurement data after noise is eliminated; calculate and evaluate the error range of the physical field through the reliability evaluation system; and identify whether an abnormal working condition occurs through the abnormal working condition identification system.

[0096] This example uses a typical high-temperature rotor as the research object. In modern high-temperature mechanical systems, high-temperature rotors are key components widely used in equipment such as steam turbines, gas turbines, and turbochargers. These rotor components are subjected to extreme thermal and mechanical loads under high-temperature, high-speed, and high-pressure operating conditions. To ensure the stability and reliability of these rotors under such harsh conditions, effective operation and maintenance are essential. Temperature monitoring plays a vital role in operation and maintenance, for example, by early detection of potential overheating issues and thermal imbalances, thereby optimizing maintenance strategies and ensuring safe operation and long-term stability of the equipment.

[0097] The high-speed rotation of the warm rotor and the high-temperature and high-pressure steam that drives the rotor to work make it extremely difficult to arrange sensors, resulting in an extremely sparse number of sensors. Often, only two or three or even one sensor point data can be obtained indirectly. Under these conditions, in order to calculate reliable high-dimensional physical fields in real time and identify abnormal working conditions, this embodiment proposes the following Figure 1 The figure shows a NIPI-HROM constrained by extremely sparse sensors, and a digital twin perception framework consisting of a reliability assessment system, a physical filtering system, and an abnormal working condition recognition system is established with NIPI-HROM as the core. Figure 3 As shown, the specific implementation of the method is mainly divided into two stages: offline data analysis and online temperature evaluation.

[0098] Offline data analysis phase:

[0099] The actual high-temperature rotor is physically modeled using CAD modeling software, and numerical mesh discretization is performed based on the finite element method. The discrete degrees of freedom are usually tens of thousands or even hundreds of thousands.

[0100] Comprehensively consider different operating conditions during the operation of high-temperature rotors, including basic operating parameters such as steam temperature rise rate, convective heat transfer coefficient between high-temperature steam and mechanical wall, and obtain test operating parameters under different loading parameters.

[0101] Conduct high-fidelity nonlinear heat conduction simulation based on finite element analysis, such as Figure 2 As shown in the figure, the third type of thermal boundary conditions in the operating conditions, namely the temperature of the steam and the convective heat transfer coefficient between the high-temperature steam and the wall, are fully considered. In each time step of the transient analysis process, the high-fidelity heat conduction matrix, heat capacity matrix and temperature field numerical results are collected in sequence after the temperature field results converge through a nonlinear iterative process.

[0102] The collected matrix data sets are the result physical quantities of finite element analysis, namely the temperature field T. Singular value decomposition is performed on the correlation matrix of the data sets to obtain their singular values ​​and eigenvectors in turn. The reduced orthogonal basis of different data sets is obtained according to the weight of the singular values, which are denoted as Φ. The reduced degrees of freedom are usually a few or dozens, which greatly reduces the degrees of freedom.

[0103] According to the specific sensor layout and discrete empirical interpolation analysis (DEIM), key physical units are selected to obtain the sensor unit index matrix Z s and the key physical unit index matrix Z p .

[0104] Online thermal assessment stage:

[0105] During high-temperature rotor operation, sensor data is input into the NIPI-HROM. A physical filtering system corrects sensor weights in real time, outputting a noise-free global temperature field. A reliability assessment system evaluates the error range of the calculated physical field, and an abnormal operating condition identification system identifies any abnormal operating conditions. Notably, the NIPI-HROM calculation process does not require the input of transient thermal boundary conditions, enabling rapid and direct reverse calculations.

[0106] The offline phase in the final embodiment involves extensive data collection, analysis, and extraction. Advanced computing hardware and parallel software design facilitate data processing and analysis, and this work serves as a prerequisite for online thermal state assessment. The online phase is not sensitive to computing hardware requirements and can be quickly implemented on a standard single-core computer.

[0107] The above describes in detail the preferred embodiments of the present invention. It should be understood that those skilled in the art can make numerous modifications and variations based on the concepts of the present invention without inventive effort. Therefore, any technical solutions that can be derived by those skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.

Claims

1. A condition monitoring and fault diagnosis method based on a digital twin perception framework, characterized in that: The process includes an offline phase and an online phase, wherein the offline phase includes the following steps: Divide the numerical calculation discrete model according to the physical structure of the industrial equipment to be tested; Based on the numerical calculation discrete model, determining the partial differential equation corresponding to the physical field, and collecting and obtaining a data set of the physical field; Performing eigendecomposition on the data set to obtain corresponding singular values ​​and eigenvectors, thereby calculating a reduced orthogonal basis of the data set; Selecting key physical units using discrete empirical interpolation analysis based on the reduced orthogonal basis to obtain a key physical unit index matrix and a key physical unit index matrix; The numerical calculation discrete model, reduced orthogonal basis, key physical unit index matrix and key physical unit index matrix are implanted into the partial differential equation of the physical field to realize real-time calculation of high-dimensional physical field, and then a reliability assessment system, a physical filtering system and an abnormal working condition identification system are respectively constructed to form a digital twin perception framework; The online phase includes the following steps: Obtaining measured data of the industrial equipment to be tested, inputting it into the digital twin perception framework, and correcting the measured data in real time through the physical filtering system to obtain noise-free measurement data; calculating and evaluating the error range of the physical field through the reliability evaluation system; and identifying whether an abnormal operating condition occurs through the abnormal operating condition recognition system; The calculation expression for feature decomposition of the data set is: Where, is the correlation matrix of the physical field data set at different times and under different parameterization conditions, The columns correspond to MM T The eigenvector of V The columns correspond to M T M The eigenvector of Σ is a singular value s 1,…, s n A diagonal matrix composed of The calculation expression of the reduced orthogonal basis is: Where, is a reduced orthogonal basis, , To reduce the orthogonal basis degrees of freedom, is the weight coefficient; The calculation expression of the key physical unit index matrix is: Where, is the key physical unit index matrix, ] is the identity matrix Middle column vectors.

2. A condition monitoring and fault diagnosis method based on a digital twin perception framework according to claim 1, characterized in that: The partial differential equation of the physical field is expressed as: Where, for t The nonlinear residual of the moment equation, u is a high-dimensional physical field, f is a nonlinear differential operator, g is the nonlinear source term, is the sensor data value, is the system parameter.

3. A condition monitoring and fault diagnosis method based on a digital twin perception framework according to claim 2, characterized in that: The partial differential equation of the physical field is iteratively solved using the Newton–Raphson iterative format until the nonlinear residual approaches 0, thereby obtaining the physical field at the current moment.

4. A condition monitoring and fault diagnosis method based on a digital twin perception framework according to claim 3, characterized in that: In the iterative solution process, k The calculation expression under the iteration step is: Where, is the increment of the physical field, and They are t Moment k The Jacobian matrix and nonlinear residual vector under the iteration step, .

5. The condition monitoring and fault diagnosis method based on the digital twin perception framework according to claim 1 is characterized in that: The calculation form obtained by implanting the numerical calculation discrete model, the reduced orthogonal basis, the key physical unit index matrix and the key physical unit index matrix into the partial differential equation of the physical field is: In the formula, the sharp corner symbol represents a low-dimensional variable, is the modal coefficient increment, and is the incomplete Jacobian matrix and nonlinear residual vector of the physical key unit set and sensor constraints, is the numerical weight of the sensor, is a reduced orthogonal basis, is the key physical unit index matrix, is the sensor unit index matrix, It is a high-dimensional physical field.

6. A condition monitoring and fault diagnosis method based on a digital twin perception framework according to claim 1, characterized in that: The reliability evaluation system uses the sum of the nonlinear residuals of the converged physical units obtained after performing real-time calculation of the high-dimensional physical field as the physical residual , taking the relative L2 norm error of the physical field as , thus based on the physical residual and the relative L2 norm error Draw a reliability characteristic curve to evaluate the error of the calculation results.

7. A condition monitoring and fault diagnosis method based on a digital twin perception framework according to claim 6, characterized in that: The calculation expression of the reliability characteristic curve is: The physical filtering system determines the stability of the sensor through the reliability evaluation system and gives greater weight to stable sensor data. , giving less weight to unstable sensor data ; The abnormal working condition identification system is based on the physical residual calculated by the reliability evaluation system The rate of change of the absolute value of the signal is compared with the preset stable rate of change range to determine whether there is a drastic change, thereby determining whether an abnormal situation occurs.

Images