A product reliability evaluation method and system based on simulation

Abstract

The application discloses a product reliability evaluation method and system based on simulation, comprising collecting monitoring data and related data of a preset product, obtaining simulation requirement data through fusion inversion of the related data and the monitoring data, taking the environmental data as a simulation background, and performing accelerated simulation on the power battery system according to the simulation requirement data and the simulation background to obtain simulation data; performing life prediction on the simulation data to obtain simulation life data, performing life monitoring based on the to-be-analyzed data to obtain actual life data, performing deviation analysis on the simulation life data and the actual life data to obtain first data; performing aging parameter comparison and analysis on the simulation data and the monitoring data to obtain second data, constructing a power battery reliability evaluation model according to the first data and the second data, and outputting an evaluation result.

Description

Technical Field

[0001] This invention relates to the field of reliability assessment, and in particular to a product reliability assessment method and system based on simulation. Background Technology

[0002] Traditional power battery reliability assessments primarily rely on experimental testing and field operation data analysis, which suffers from limitations such as long testing cycles, high costs, and incomplete coverage of operating conditions. With the increasing demands for battery life and safety from new energy vehicles and energy storage systems, existing technologies are insufficient to meet the precise assessment needs under complex coupled scenarios. Early methods often employed accelerated aging experiments combined with statistical models, estimating lifespan through single stress acceleration factors such as temperature and rate, but these methods cannot effectively reflect the nonlinear degradation characteristics under multi-field coupling. While numerical simulation techniques developed in recent years can construct electrochemical-thermal-mechanical coupled models, they are limited by missing material characteristic parameters and simplified boundary conditions such as the dynamic response of the BMS strategy, leading to systematic deviations between simulation results and actual operating conditions.

[0003] Existing technologies face three key bottlenecks: First, insufficient fusion of multi-source heterogeneous data, with non-electrical indicators such as environmental data and manufacturing process parameters not effectively incorporated into the evaluation system; second, the aging mechanism modeling is dimensionally limited, with existing research focusing primarily on independent analyses of chemical degradation or mechanical failure, lacking collaborative simulations of electro-thermal-mechanical multi-field coupled failures; and third, a disconnect between real-time diagnosis and prediction, with offline models based on historical data struggling to respond promptly to dynamic changes in operating conditions. Currently, data-driven lifetime prediction methods have been proposed, but the closed-loop verification problem between simulation and measured data remains unresolved; while electrochemical impedance spectroscopy is used to establish aging models, characterization methods for interface failures induced by mechanical stress are lacking.

[0004] Therefore, there is an urgent need for a new reliability assessment method for power battery products based on simulation, to improve the accuracy and full-cycle integrity of reliability assessment of power battery systems. Summary of the Invention

[0005] The purpose of this invention is to provide a product reliability assessment method based on simulation.

[0006] To achieve the above objectives, the present invention is implemented according to the following technical solution:

[0007] This invention includes the following steps:

[0008] The system collects monitoring data and related data of a preset product, and preprocesses the monitoring data and related data; the related data includes environmental data, historical operating data, manufacturing process data, and design parameters of the power battery; the monitoring data includes historical data and data to be evaluated; the historical data includes electrical parameters, control signals, thermal parameters, and mechanical parameters.

[0009] Simulation requirement data is obtained by fusing and inverting the relevant data and the monitoring data, and the environmental data is used as the simulation background. The power battery system is then subjected to accelerated simulation based on the simulation requirement data and the simulation background to obtain simulation data. The simulation requirement data includes electrode material characteristics, separator parameters, single cell data, module data, electrothermal coupling data, electromechanical coupling data, operating condition and environmental data, chemical degradation parameters, mechanical failure data, BMS strategy parameters, fault response logic, in-situ test data, electrochemical test data, and accelerated aging data.

[0010] The simulated life data is obtained by performing life prediction on the simulated data, the actual life data is obtained by performing life monitoring based on the data to be analyzed, and the first data is obtained by performing deviation analysis based on the simulated life data and the actual life data.

[0011] Second data is obtained by comparing and analyzing aging parameters of the simulated data and the monitoring data. A power battery reliability assessment model is constructed based on the first data and the second data. The data to be assessed is input into the power battery reliability assessment model, and the assessment result is output.

[0012] Furthermore, the method for obtaining simulation requirement data by fusing and inverting the relevant data and the monitoring data includes:

[0013] Feature extraction is performed on relevant data and monitoring data to obtain inversion demand features; the inversion features include physical features, statistical time-series features, and correlation features; physical features include electrical parameters, thermal parameters, and mechanical parameters; statistical time-series features include mean, variance, kurtosis, and time-series characteristics; among them, correlation features represent mapping process parameters to material properties;

[0014] Based on the characteristics of the inversion demand, relevant data, and monitoring data, multi-source data fusion is performed, and a joint probability distribution is defined, expressed as:

[0015]

[0016] The inversion demand characteristics are: The monitoring data is h mom The environmental data is h env Historical data is h his The process data is h pr Given historical data h his and process data h pr The prior distribution of time-inverted demand characteristics is Given the inversion requirement characteristics, the probability of monitoring data is Given the inversion demand characteristics, the probability of environmental data is: Given monitoring data, environmental data, historical data, and process data, the joint probability distribution for retrieving demand characteristics is as follows:

[0017] Construct an inversion model based on Markov chain Monte Carlo, and define an objective function based on the minimum difference between the monitored data and the simulation output. The expression is as follows:

[0018]

[0019] Where the objective function is Battery simulation based on inverted demand characteristics and environmental data is used for...

[0020] Given a loss function, the expression is:

[0021]

[0022] The loss function is L fy The predicted voltage is V pre The monitored voltage is V mom The physical constraint loss is L phy The physical weighting coefficient is α;

[0023] Input the inversion demand characteristics and monitoring data into the inversion model to obtain the predicted voltage and predicted temperature, and use the predicted voltage and predicted temperature as the simulation demand data.

[0024] Furthermore, the method for obtaining simulation data by performing accelerated simulation on the power battery system includes:

[0025] Monte Carlo methods are used to randomly perturb the parameters in the simulation requirement data to obtain uncertainty data;

[0026] Based on the simulation requirements data and simulation background, an electrochemical-thermomechanical coupling model is constructed, in which the expressions for solid-phase diffusion, electrolyte potential, energy conservation, and strain-stress relationships are:

[0027]

[0028] σ=F·γ Li +β·ΔT

[0029] The concentration of the solid phase liquid is H. S The time is t, and the solid diffusion coefficient is U. S The gradient operator is The current density is i, the electrolyte conductivity is η, and the potential is... The gas constant is W, the temperature is T, and the transport number is t. + The electrolyte concentration is H e The density is ρ and the specific heat capacity is D.u The thermal conductivity is z, the efficiency is θ, and the heat generated by the current concentration and efficiency is M. gen (i,θ), the environmental heat source is M env The stress is σ, the elastic modulus is F, and the strain caused by the lithium particles is γ. Li The temperature change is ΔT, the Poisson's ratio is β, the adjustment factor is ζ, and the critical temperature is T. crit ;

[0030] The electrochemical-thermomechanical coupling model is reduced in order by intrinsic orthogonal decomposition and accelerated by dynamic time step strategy; the dynamic time step strategy is to increase compensation in non-critical stages and adaptively encrypt in critical stages.

[0031] The state of the power battery includes its current remaining charge, temperature, and aging stage, which are accelerated by time compression factor and current rate adjustment, expressed as:

[0032]

[0033] Where the time scaling factor is The initial temperature is T1, the target temperature is T2, the gas constant is B, and the activation energy is S. a ;

[0034] The current multiplier is adjusted to twice the initial value, and the mean square error between the target voltage and the simulated voltage is used as the reward function.

[0035] Risk values ​​are obtained by performing risk assessment based on simulation requirements data. A fault probability map is constructed based on the risk values. When the risk value is greater than 0.673, fault seeds are injected. The injection strategy is dynamically adjusted by comparing monitoring data with data twins.

[0036] A virtual simulation environment is used to perform simulation requirements data after adjustment and injection. The near-end policy optimization algorithm is trained in the virtual simulation environment until the reward function value reaches the minimum. The dynamic time step is output, and the simulation data is given based on the dynamic time step.

[0037] Furthermore, a method for obtaining simulated lifetime data by performing lifetime prediction on the simulated data includes:

[0038] Health factors related to battery health status were extracted, and the correlation coefficient was calculated based on the causal relationship between battery life and health factors using the English causality test. Health factors with a correlation coefficient greater than 0.831 were used as covariates. Covariates included magnetothermal coupling factor, coil temperature rise rate, and bearing current harmonic distortion rate.

[0039] Obtain the time step of the accelerated simulation, input the simulation data, covariates, and time step into the lifetime prediction model, and calculate the output lifetime at the current time step:

[0040]

[0041] The lifespan model parameters are: The output lifetime of the b-th time step is k b,t The b-th covariate at the current time step is x. b,t The lifetime of the b-th observation at time step t-1 is f. b,t-1 The lifetime of the b-th output at time t-1 is k. b,t-1 ,

[0042] A dual-channel time attention model is introduced to focus on key time points. All time steps of the network are input into the dual-channel time attention model to obtain the deep relationship between lifetime and time steps. The dual channels are the hardware degradation channel and the transient time channel.

[0043] By weighting the outputs at different time steps using a weighted sequence, the cumulative sum is calculated, and the final output is obtained.

[0044]

[0045] The attention weight for the hardware degradation channel at time t is φ1. t The transient time channel attention weight at time t is φ2. t The output lifetime at time t is k t The cumulative sum of time t is The start time of the time step is t. o The deadline for the time step is T. s ;

[0046] Construct a mixture likelihood model using Gaussian and Weibull distributions, and calculate the likelihood function:

[0047] P t =ω·N(μ,δ 2 )+(1-ω)Wb(y,ξ)

[0048] The mixing coefficient is ω, the shape parameter is y, the scale parameter is ξ, and the standard deviation is μ. The Weibull distributions of the shape parameter y and the scale parameter ξ are Wb(y,λ), with mean μ and variance δ. 2 The normal distribution is N(μ,δ) 2 The mixture likelihood function for time t is P. t ;

[0049] Introducing regularization, we regularize the loss of the lifetime prediction model, expressed as:

[0050]

[0051] Where the loss function is The initial regularization strength is λo The loss functions for predicted lifetime and actual lifetime are: The loss function of the lifetime prediction model is The parameter vector of the a-th covariate is c a The number of parameter vectors is N a The parameter sparsity penalty is λ. o ·e -0.01t The weight decays to

[0052] The residuals are corrected using correction factors, and the simulated remaining useful life is calculated:

[0053]

[0054] The simulated remaining lifespan of the power battery is: The given parameter for time t is Given parameters The likelihood function is The correction factor is The start time is t o The termination time is t sm The remaining service life is

[0055] The simulated remaining useful life will be output as simulated life data.

[0056] Furthermore, the method for obtaining the first data by performing deviation analysis based on the simulated lifetime data and the actual lifetime data includes:

[0057] Calculate the mean square error between simulated life data and actual life data. When the mean square error is less than 0.05, it is considered acceptable error; when the mean square error is greater than 0.05 but less than 0.156, it is considered error requiring parameter calibration; when the mean square error is greater than 0.156, it is considered error due to model structural defects.

[0058] Sensitive parameters are obtained by performing multi-dimensional correlation analysis on the mean square error; the multi-dimensional analysis includes time-domain cumulative bias, frequency-domain characteristics, and residual autocorrelation.

[0059] The problem data is obtained by jointly analyzing sensitive parameters in the time and frequency domains. The joint analysis includes: using cumulative sums and control charts to identify the systemic lifetime prediction model offset by the cumulative deviation in the time domain; using windowed piecewise average power spectral density estimation to identify unmodeled periodic perturbations in the frequency domain; and using residual autocorrelation to identify dynamic response delay.

[0060] Simulate the problem data to obtain simulated problem data, and calculate the first data:

[0061]

[0062] The first data point of time t is d. t The number of problem data is N w The time constant is D, and the regularization coefficient is The number of sensitive parameters is N v The vth sensitive parameter is The deviation of the sensitive parameter from the prior value is The data for the w-th simulation problem is ds w The data for the wth question is dr w Problem data dr w The time of occurrence is t w The time is t.

[0063] Furthermore, the method for obtaining the second data by comparing and analyzing the aging parameters of the simulated data and the monitoring data includes:

[0064] A graph neural network is used to construct the spatial correlation information of the internal structure of the power battery. Short-time Fourier transform and wavelet transform are used to extract aging parameters from simulated data and monitoring data at multiple scales. The multiple scales include macroscopic, microscopic and mesoscopic. The aging features include capacity decay rate, relaxation time distribution and internal resistance increment slope. The aging parameters include simulated aging parameters and actual aging parameters.

[0065] The aging parameters are spatially mapped using a kernel function, expressed as follows:

[0066]

[0067] Where the x-axis of the aging parameter is d, the y-axis of the aging parameter is g, and the kernel function of the aging parameter coordinate (d, g) is: The standard deviation of aging parameters is The parameters affecting the adjustment are The cycle period is τ;

[0068] A deep feature matching mechanism is introduced to measure aging similarity, and aging similarity is calculated based on dynamic weights:

[0069]

[0070] The aging similarity at time t is The dynamic weight of the r-th aging parameter at time t is: The dynamic weight of the r-th aging parameter at time t+1 is: The r-th simulated aging parameter is The r-th actual aging parameter is The learning rate is χ, and the gradient of the r-th aging similarity is... The gradient of the p-th aging similarity is The dynamic weight of the p-th aging parameter is The number of aging parameters is N r ;

[0071] Simulated data and monitoring data are paired based on aging similarity to obtain paired data, and the standard deviation of the paired data is used as the second data.

[0072] Furthermore, the method for constructing a power battery reliability assessment model based on the first data and the second data includes:

[0073] The actual remaining battery life is obtained based on monitoring data. An objective function is constructed based on the first and second data points, and its expression is:

[0074]

[0075] Where the objective function for the j-th evaluation at time t is: Simulated remaining service life is Actual remaining service life is The loss functions for simulated remaining useful life and actual remaining useful life are as follows: The first data point for the j-th evaluation at time t is d. t (j), the second data of the j-th evaluation at time t is ψ t (j);

[0076] The reliability assessment model for power batteries includes self-encoders, long short-term memory networks, and artificial neural network algorithms;

[0077] An autoencoder encodes input data into a low-dimensional representation, decodes and restores it, learns the inherent characteristics of the input data, and extracts key features based on their correlation with product reliability.

[0078] Long Short-Term Memory (LSTM) networks selectively remember and forget key long-term and short-term information through memory units and gating mechanisms, and obtain temporal characteristics by performing temporal characteristic analysis on long-term and short-term information.

[0079] Artificial neural network algorithms are based on objective functions. Through network training and learning, they can uncover the potential relationship between time-series characteristics and the reliability of power battery products, and obtain reliability assessment values ​​based on the potential relationship.

[0080] Secondly, a product reliability assessment system based on simulation includes:

[0081] Data acquisition module: used to collect monitoring data and related data of preset products, and preprocess the monitoring data and related data; the related data includes environmental data, historical operating data, manufacturing process data and design parameters of the power battery; the monitoring data includes historical data and data to be evaluated; the historical data includes electrical parameters, control signals, thermal parameters and mechanical parameters;

[0082] The simulation module is used to obtain simulation requirement data by fusing and inverting the relevant data and the monitoring data, using the environmental data as the simulation background, and performing accelerated simulation on the power battery system according to the simulation requirement data and the simulation background to obtain simulation data. The simulation requirement data includes electrode material characteristics, separator parameters, single cell data, module data, electrothermal coupling data, electromechanical coupling data, operating condition and environmental data, chemical degradation parameters, mechanical failure data, BMS strategy parameters, fault response logic, in-situ test data, and electrochemical test data.

[0083] Prediction Deviation Analysis Module: Used to perform lifetime prediction on the simulation data to obtain simulated lifetime data, perform lifetime monitoring based on the data to be analyzed to obtain actual lifetime data, and perform deviation analysis based on the simulated lifetime data and the actual lifetime data to obtain first data;

[0084] Modeling and evaluation module: used to obtain second data by comparing and analyzing aging parameters of the simulation data and the monitoring data, construct a power battery reliability evaluation model based on the first data and the second data, input the data to be evaluated into the power battery reliability evaluation model, and output the evaluation results.

[0085] The beneficial effects of this invention are:

[0086] This invention is a product reliability assessment method and system based on simulation. Compared with the prior art, this invention has the following technical advantages:

[0087] This invention improves the accuracy of reliability assessment for power battery products through preprocessing, fusion inversion, accelerated simulation, lifetime prediction, deviation analysis, aging parameter comparison analysis, and model building steps. This enhances the precision of reliability assessment, optimizes the reliability assessment of multiple power battery products, significantly saves resources, improves work efficiency, and enables intelligent reliability assessment of power battery products. It provides real-time accelerated simulation and deviation analysis for reliability assessment, which is of great significance for power battery product reliability assessment. It can adapt to reliability assessment requirements of power battery products with different standards and needs, demonstrating a certain degree of universality. Attached Figure Description

[0088] Figure 1 This is a flowchart illustrating the steps of a product reliability assessment method based on simulation according to the present invention. Detailed Implementation

[0089] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0090] The present invention provides a product reliability assessment method and system based on simulation, comprising the following steps:

[0091] like Figure 1 As shown, this embodiment includes the following steps:

[0092] The system collects monitoring data and related data of a preset product, and preprocesses the monitoring data and related data; the related data includes environmental data, historical operating data, manufacturing process data, and design parameters of the power battery; the monitoring data includes historical data and data to be evaluated; the historical data includes electrical parameters, control signals, thermal parameters, and mechanical parameters.

[0093] In the actual evaluation, a certain model of electric vehicle power battery was used as the research object. Historical data included electrical parameters such as average voltage of 3.7V, fluctuation range of 3.6-3.8V, and maximum charging and discharging current of 100A; thermal parameters such as surface temperature of 25-35℃ during normal driving; and mechanical parameters such as the average vibration frequency of the battery module of 5Hz.

[0094] Relevant data includes environmental data recording driving ambient temperature of 20-30℃, historical operation data including a total mileage of 20,000 kilometers and 500 charge-discharge cycles in the past year, manufacturing process data showing an electrode material coating thickness of 0.1mm, and design parameters including a battery capacity of 50Ah.

[0095] Simulation requirement data is obtained by fusing and inverting the relevant data and the monitoring data, and the environmental data is used as the simulation background. The power battery system is then subjected to accelerated simulation based on the simulation requirement data and the simulation background to obtain simulation data. The simulation requirement data includes electrode material characteristics, separator parameters, single cell data, module data, electrothermal coupling data, electromechanical coupling data, operating condition and environmental data, chemical degradation parameters, mechanical failure data, BMS strategy parameters, fault response logic, in-situ test data, electrochemical test data, and accelerated aging data.

[0096] In the actual evaluation, physical characteristics are obtained based on electrical, thermal, and mechanical parameters; statistical time-series characteristics are calculated, with the voltage mean being 3.7V and variance being 0.01; inversion calculations are performed by defining a joint probability distribution through multi-source data fusion; simulation requirements include the ion diffusion coefficient of the electrode material.

[0097] The initial temperature is set at 25℃, and the target temperature is set at 35℃.

[0098] The simulated life data is obtained by performing life prediction on the simulated data, the actual life data is obtained by performing life monitoring based on the data to be analyzed, and the first data is obtained by performing deviation analysis based on the simulated life data and the actual life data.

[0099] In the actual assessment, the first data point was 0.372;

[0100] The second data is obtained by comparing and analyzing the aging parameters of the simulated data and the monitoring data. A power battery reliability assessment model is constructed based on the first data and the second data. The data to be assessed is input into the power battery reliability assessment model, and the assessment result is output.

[0101] In the actual assessment, the second data was 0.356, and the assessment result was 0.709.

[0102] In this embodiment, the method for obtaining simulation requirement data by fusing and inverting the relevant data and the monitoring data includes:

[0103] Feature extraction is performed on relevant data and monitoring data to obtain inversion demand features; the inversion features include physical features, statistical time-series features, and correlation features; physical features include electrical parameters, thermal parameters, and mechanical parameters; statistical time-series features include mean, variance, kurtosis, and time-series characteristics; among them, correlation features represent mapping process parameters to material properties;

[0104] Based on the characteristics of the inversion demand, relevant data, and monitoring data, multi-source data fusion is performed, and a joint probability distribution is defined, expressed as:

[0105]

[0106] The inversion demand characteristics are: The monitoring data is h mom The environmental data is h env Historical data is h his The process data is h pr Given historical data h his and process data h pr The prior distribution of time-inverted demand characteristics is Given the inversion requirement characteristics, the probability of monitoring data is Given the inversion demand characteristics, the probability of environmental data is: Given monitoring data, environmental data, historical data, and process data, the joint probability distribution for retrieving demand characteristics is as follows:

[0107] Construct an inversion model based on Markov chain Monte Carlo, and define an objective function based on the minimum difference between the monitored data and the simulation output. The expression is as follows:

[0108]

[0109] Where the objective function is Battery simulation based on inverted demand characteristics and environmental data is used for...

[0110] Given a loss function, the expression is:

[0111] L fy =||V pre -V mom || 2 +α·L phy

[0112] The loss function is L fy The predicted voltage is V pre The monitored voltage is V mom The physical constraint loss is L phy The physical weighting coefficient is α;

[0113] Input the inversion demand characteristics and monitoring data into the inversion model to obtain the predicted voltage and predicted temperature, and use the predicted voltage and predicted temperature as the simulation demand data.

[0114] In this embodiment, the method for obtaining simulation data by performing accelerated simulation of the power battery system includes:

[0115] Monte Carlo methods are used to randomly perturb the parameters in the simulation requirement data to obtain uncertainty data;

[0116] Based on the simulation requirements data and simulation background, an electrochemical-thermomechanical coupling model is constructed, in which the expressions for solid-phase diffusion, electrolyte potential, energy conservation, and strain-stress relationships are:

[0117]

[0118] σ=F·γ Li +β·ΔT

[0119] The concentration of the solid phase liquid is H. S The time is t, and the solid diffusion coefficient is U. S The gradient operator is The current density is i, the electrolyte conductivity is η, and the potential is... The gas constant is W, the temperature is T, and the transport number is t. + The electrolyte concentration is H e The density is ρ and the specific heat capacity is D. u The thermal conductivity is z, the efficiency is θ, and the heat generated by the current concentration and efficiency is M. gen (i,θ), the environmental heat source is M env The stress is σ, the elastic modulus is F, and the strain caused by the lithium particles is γ. Li The temperature change is ΔT, the Poisson's ratio is β, the adjustment factor is ζ, and the critical temperature is T. crit ;

[0120] The electrochemical-thermomechanical coupling model is reduced in order by intrinsic orthogonal decomposition and accelerated by dynamic time step strategy; the dynamic time step strategy is to increase compensation in non-critical stages and adaptively encrypt in critical stages.

[0121] The state of the power battery includes its current remaining charge, temperature, and aging stage, which are accelerated by time compression factor and current rate adjustment, expressed as:

[0122]

[0123] Where the time scaling factor is The initial temperature is T1, the target temperature is T2, the gas constant is B, and the activation energy is S. a ;

[0124] The current multiplier is adjusted to twice the initial value, and the mean square error between the target voltage and the simulated voltage is used as the reward function.

[0125] Risk values ​​are obtained by performing risk assessment based on simulation requirements data. A fault probability map is constructed based on the risk values. When the risk value is greater than 0.673, fault seeds are injected. The injection strategy is dynamically adjusted by comparing monitoring data with data twins.

[0126] A virtual simulation environment is used to perform simulation requirements data after adjustment and injection. The near-end policy optimization algorithm is trained in the virtual simulation environment until the reward function value reaches the minimum. The dynamic time step is output, and the simulation data is given based on the dynamic time step.

[0127] In this embodiment, the method for obtaining simulated lifetime data by performing lifetime prediction on the simulated data includes:

[0128] Health factors related to battery health status were extracted, and the correlation coefficient was calculated based on the causal relationship between battery life and health factors using the English causality test. Health factors with a correlation coefficient greater than 0.831 were used as covariates. Covariates included magnetothermal coupling factor, coil temperature rise rate, and bearing current harmonic distortion rate.

[0129] Obtain the time step of the accelerated simulation, input the simulation data, covariates, and time step into the lifetime prediction model, and calculate the output lifetime at the current time step:

[0130]

[0131] The lifespan model parameters are: The output lifetime of the b-th time step is k b,t The b-th covariate at the current time step is x. b,t The lifetime of the b-th observation at time step t-1 is f. b,t-1 The lifetime of the b-th output at time t-1 is k. b,t-1 ,

[0132] A dual-channel time attention model is introduced to focus on key time points. All time steps of the network are input into the dual-channel time attention model to obtain the deep relationship between lifetime and time steps. The dual channels are the hardware degradation channel and the transient time channel.

[0133] By weighting the outputs at different time steps using a weighted sequence, the cumulative sum is calculated, and the final output is obtained.

[0134]

[0135] The attention weight for the hardware degradation channel at time t is φ1. t The transient time channel attention weight at time t is φ2. t The output lifetime at time t is k t The cumulative sum of time t is The start time of the time step is t. o The deadline for the time step is T. s ;

[0136] Construct a mixture likelihood model using Gaussian and Weibull distributions, and calculate the likelihood function:

[0137] P t =ω·N(μ,δ 2 )+(1-ω)Wb(y,ξ)

[0138] The mixing coefficient is ω, the shape parameter is y, the scale parameter is ξ, and the standard deviation is μ. The Weibull distributions of the shape parameter y and the scale parameter ξ are Wb(y,λ), with mean μ and variance δ. 2 The normal distribution is N(μ,δ)2 The mixture likelihood function for time t is P. t ;

[0139] Introducing regularization, we regularize the loss of the lifetime prediction model, expressed as:

[0140]

[0141] Where the loss function is The initial regularization strength is λ o The loss functions for predicted lifetime and actual lifetime are: The loss function of the lifetime prediction model is The parameter vector of the a-th covariate is c a The number of parameter vectors is N a The parameter sparsity penalty is λ. o ·e -0.01t The weight decays to

[0142] The residuals are corrected using correction factors, and the simulated remaining useful life is calculated:

[0143]

[0144] The simulated remaining lifespan of the power battery is: The given parameter for time t is Given parameters The likelihood function is The correction factor is The start time is t o The termination time is t sm The remaining service life is

[0145] The simulated remaining useful life will be output as simulated life data.

[0146] In this embodiment, the method for obtaining first data by performing deviation analysis based on the simulated lifetime data and the actual lifetime data includes:

[0147] Calculate the mean square error between simulated life data and actual life data. When the mean square error is less than 0.05, it is considered acceptable error; when the mean square error is greater than 0.05 but less than 0.156, it is considered error requiring parameter calibration; when the mean square error is greater than 0.156, it is considered error due to model structural defects.

[0148] Sensitive parameters are obtained by performing multi-dimensional correlation analysis on the mean square error; the multi-dimensional analysis includes time-domain cumulative bias, frequency-domain characteristics, and residual autocorrelation.

[0149] The problem data is obtained by jointly analyzing sensitive parameters in the time and frequency domains. The joint analysis includes: using cumulative sums and control charts to identify the systemic lifetime prediction model offset by the cumulative deviation in the time domain; using windowed piecewise average power spectral density estimation to identify unmodeled periodic perturbations in the frequency domain; and using residual autocorrelation to identify dynamic response delay.

[0150] Simulate the problem data to obtain simulated problem data, and calculate the first data:

[0151]

[0152] The first data point of time t is d. t The number of problem data is N w The time constant is D, and the regularization coefficient is The number of sensitive parameters is N v The vth sensitive parameter is The deviation of the sensitive parameter from the prior value is The data for the w-th simulation problem is ds w The data for the wth question is dr w Problem data dr w The time of occurrence is t w The time is t.

[0153] In this embodiment, the method for obtaining second data by comparing and analyzing aging parameters of the simulated data and the monitoring data includes:

[0154] A graph neural network is used to construct the spatial correlation information of the internal structure of the power battery. Short-time Fourier transform and wavelet transform are used to extract aging parameters from simulated data and monitoring data at multiple scales. The multiple scales include macroscopic, microscopic and mesoscopic. The aging features include capacity decay rate, relaxation time distribution and internal resistance increment slope. The aging parameters include simulated aging parameters and actual aging parameters.

[0155] The aging parameters are spatially mapped using a kernel function, expressed as follows:

[0156]

[0157] Where the x-axis of the aging parameter is d, the y-axis of the aging parameter is g, and the kernel function of the aging parameter coordinate (d, g) is: The standard deviation of aging parameters is The parameters affecting the adjustment are The cycle period is τ;

[0158] A deep feature matching mechanism is introduced to measure aging similarity, and aging similarity is calculated based on dynamic weights:

[0159]

[0160] The aging similarity at time t is The dynamic weight of the r-th aging parameter at time t is: The dynamic weight of the r-th aging parameter at time t+1 is: The r-th simulated aging parameter is The r-th actual aging parameter is The learning rate is χ, and the gradient of the r-th aging similarity is... The gradient of the p-th aging similarity is The dynamic weight of the p-th aging parameter is The number of aging parameters is N r ;

[0161] Simulated data and monitoring data are paired based on aging similarity to obtain paired data, and the standard deviation of the paired data is used as the second data.

[0162] In this embodiment, the method for constructing a power battery reliability assessment model based on the first data and the second data includes:

[0163] The actual remaining battery life is obtained based on monitoring data. An objective function is constructed based on the first and second data points, and its expression is:

[0164]

[0165] Where the objective function for the j-th evaluation at time t is: Simulated remaining service life is Actual remaining service life is The loss functions for simulated remaining useful life and actual remaining useful life are as follows: The first data point for the j-th evaluation at time t is d. t (j), the second data of the j-th evaluation at time t is ψ t (j);

[0166] The reliability assessment model for power batteries includes self-encoders, long short-term memory networks, and artificial neural network algorithms;

[0167] An autoencoder encodes input data into a low-dimensional representation, decodes and restores it, learns the inherent characteristics of the input data, and extracts key features based on their correlation with product reliability.

[0168] Long Short-Term Memory (LSTM) networks selectively remember and forget key long-term and short-term information through memory units and gating mechanisms, and obtain temporal characteristics by performing temporal characteristic analysis on long-term and short-term information.

[0169] Artificial neural network algorithms are based on objective functions. Through network training and learning, they can uncover the potential relationship between time-series characteristics and the reliability of power battery products, and obtain reliability assessment values ​​based on the potential relationship.

[0170] Secondly, a product reliability assessment system based on simulation includes:

[0171] Data acquisition module: used to collect monitoring data and related data of preset products, and preprocess the monitoring data and related data; the related data includes environmental data, historical operating data, manufacturing process data and design parameters of the power battery; the monitoring data includes historical data and data to be evaluated; the historical data includes electrical parameters, control signals, thermal parameters and mechanical parameters;

[0172] The simulation module is used to obtain simulation requirement data by fusing and inverting the relevant data and the monitoring data, using the environmental data as the simulation background, and performing accelerated simulation on the power battery system according to the simulation requirement data and the simulation background to obtain simulation data. The simulation requirement data includes electrode material characteristics, separator parameters, single cell data, module data, electrothermal coupling data, electromechanical coupling data, operating condition and environmental data, chemical degradation parameters, mechanical failure data, BMS strategy parameters, fault response logic, in-situ test data, and electrochemical test data.

[0173] Prediction Deviation Analysis Module: Used to perform lifetime prediction on the simulation data to obtain simulated lifetime data, perform lifetime monitoring based on the data to be analyzed to obtain actual lifetime data, and perform deviation analysis based on the simulated lifetime data and the actual lifetime data to obtain first data;

[0174] Modeling and evaluation module: used to obtain second data by comparing and analyzing aging parameters of the simulation data and the monitoring data, construct a power battery reliability evaluation model based on the first data and the second data, input the data to be evaluated into the power battery reliability evaluation model, and output the evaluation results.

[0175] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A product reliability assessment method based on simulation, characterized in that, Includes the following steps: The system collects monitoring data and related data of a preset product, and preprocesses the monitoring data and related data; the related data includes environmental data, historical operating data, manufacturing process data, and design parameters of the power battery; the monitoring data includes historical data and data to be evaluated; the historical data includes electrical parameters, control signals, thermal parameters, and mechanical parameters. Simulation requirement data is obtained by fusing and inverting the relevant data and the monitoring data, and the environmental data is used as the simulation background. The power battery system is then subjected to accelerated simulation based on the simulation requirement data and the simulation background to obtain simulation data. The simulation requirement data includes electrode material characteristics, separator parameters, single cell data, module data, electrothermal coupling data, electromechanical coupling data, operating condition and environmental data, chemical degradation parameters, mechanical failure data, BMS strategy parameters, fault response logic, in-situ test data, electrochemical test data, and accelerated aging data. The simulated life data is obtained by performing life prediction on the simulated data, the actual life data is obtained by performing life monitoring based on the data to be analyzed, and the first data is obtained by performing deviation analysis based on the simulated life data and the actual life data. The second data is obtained by comparing and analyzing the aging parameters of the simulated data and the monitoring data. The first data and the second data are used to construct a power battery reliability assessment model. The data to be assessed is input into the power battery reliability assessment model, and the assessment result is output.

2. The product reliability assessment method based on simulation according to claim 1, characterized in that, A method for obtaining simulation requirement data by fusing and inverting the relevant data and the monitoring data includes: Feature extraction is performed on relevant data and monitoring data to obtain inversion demand features; the inversion features include physical features, statistical time-series features, and correlation features; physical features include electrical parameters, thermal parameters, and mechanical parameters; statistical time-series features include mean, variance, kurtosis, and time-series characteristics; among them, correlation features represent mapping process parameters to material properties; Based on the characteristics of the inversion demand, relevant data, and monitoring data, multi-source data fusion is performed, and a joint probability distribution is defined, expressed as: ； The inversion demand characteristics are: The monitoring data is Environmental data is Historical data is The process data is Given historical data and process data The prior distribution of time-inverted demand characteristics is Given the inversion requirement characteristics, the probability of monitoring data is: Given the inversion demand characteristics, the probability of environmental data is: Given monitoring data, environmental data, historical data, and process data, the joint probability distribution of the inverted demand characteristics is as follows: ; Construct an inversion model based on Markov chain Monte Carlo, and define an objective function based on the minimum difference between the monitored data and the simulation output. The expression is as follows: ； Where the objective function is Battery simulation based on inverted demand characteristics and environmental data is used for ; Given a loss function, the expression is: ； Where the loss function is The predicted voltage is The monitored voltage is The physical constraint loss is The physical weighting coefficient is ; Input the inversion demand characteristics and monitoring data into the inversion model to obtain the predicted voltage and predicted temperature, and use the predicted voltage and predicted temperature as the simulation demand data.

3. The product reliability assessment method based on simulation according to claim 1, characterized in that, A method for obtaining simulation data by performing accelerated simulation on the power battery system includes: Monte Carlo methods are used to randomly perturb the parameters in the simulation requirement data to obtain uncertainty data; Based on the simulation requirements data and simulation background, an electrochemical-thermomechanical coupling model is constructed, in which the expressions for solid-phase diffusion, electrolyte potential, energy conservation, and strain-stress relationships are: ； ； ； ； The concentration of the solid phase liquid is The time is t, and the solid diffusion coefficient is The gradient operator is The current density is i, and the electrolyte conductivity is The potential is The gas constant is W, the temperature is T, and the transport number is... The electrolyte concentration is The density is Specific heat capacity is Thermal conductivity is Efficiency is The heat generated by current concentration and efficiency is The environmental heat source is The stress is The elastic modulus is F, and the strain caused by lithium particles is Temperature change is Poisson's ratio is The adjustment factor is The critical temperature is ; The electrochemical-thermomechanical coupling model is reduced in order by intrinsic orthogonal decomposition and accelerated by dynamic time step strategy; the dynamic time step strategy is to increase compensation in non-critical stages and adaptively encrypt in critical stages. The state of the power battery includes its current remaining charge, temperature, and aging stage, which are accelerated by time compression factor and current rate adjustment, expressed as: ； Where the time scaling factor is The initial temperature is The target temperature is The gas constant is B, and the activation energy is... ; The current multiplier is adjusted to twice the initial value, and the mean square error between the target voltage and the simulated voltage is used as the reward function. Risk values ​​are obtained by performing risk assessment based on simulation requirements data. A fault probability map is constructed based on the risk values. When the risk value is greater than 0.673, fault seeds are injected. The injection strategy is dynamically adjusted by comparing monitoring data with data twins. A virtual simulation environment is used to perform simulation requirements data after adjustment and injection. The near-end policy optimization algorithm is trained in the virtual simulation environment until the reward function value reaches the minimum. The dynamic time step is output, and the simulation data is given based on the dynamic time step.

4. The product reliability assessment method based on simulation according to claim 1, characterized in that, A method for obtaining simulated lifetime data by performing lifetime prediction on the simulated data includes: Health factors related to battery health status were extracted, and the correlation coefficient was calculated based on the causal relationship between battery life and health factors using the English causality test. Health factors with a correlation coefficient greater than 0.831 were used as covariates. Covariates included magnetothermal coupling factor, coil temperature rise rate, and bearing current harmonic distortion rate. Obtain the time step of the accelerated simulation, input the simulation data, covariates, and time step into the lifetime prediction model, and calculate the output lifetime at the current time step: ； The lifespan model parameters are: The output lifetime of the b-th time step is The b-th covariate at the current time step is The lifetime of the b-th observation at time step t-1 is The lifetime of the b-th output at time t-1 is , A dual-channel time attention model is introduced to focus on key time points. All time steps of the network are input into the dual-channel time attention model to obtain the deep relationship between lifetime and time steps. The dual channels are the hardware degradation channel and the transient time channel. By weighting the outputs at different time steps using a weighted sequence, the cumulative sum is calculated, and the final output is obtained. ； ； The attention weight for the hardware degradation channel at time t is... The transient time channel attention weights at time t are: The output lifetime at time t is The cumulative sum of time t is The start time of the time step is The deadline for the time step is ; Construct a mixture likelihood model using Gaussian and Weibull distributions, and calculate the likelihood function: ； Where the mixing coefficient is The shape parameter is y, and the scale parameter is... The standard deviation is Shape parameter y, scale parameter The Weibull distribution is mean ,variance normal distribution The mixture likelihood function for time t is: ; Introducing regularization, we regularize the loss of the lifetime prediction model, expressed as: ； The loss functions for predicted lifetime and actual lifetime are: The loss function of the lifetime prediction model is The parameter vector of the a-th covariate is The number of parameter vectors is The parameter sparsity penalty is The weight decays to ; The residuals are corrected using correction factors, and the simulated remaining useful life is calculated: ； The simulated remaining lifespan of the power battery is: The given parameter for time t is Given parameters The likelihood function is The correction factor is The start time is The end time is ; The simulated remaining useful life will be output as simulated life data.

5. The product reliability assessment method based on simulation according to claim 1, characterized in that, A method for obtaining first data by performing deviation analysis based on the simulated lifetime data and the actual lifetime data includes: Calculate the mean square error between simulated life data and actual life data. When the mean square error is less than 0.05, it is considered acceptable error; when the mean square error is greater than 0.05 but less than 0.156, it is considered error requiring parameter calibration; when the mean square error is greater than 0.156, it is considered error due to model structural defects. Sensitive parameters are obtained by performing multi-dimensional correlation analysis on the mean square error; the multi-dimensional analysis includes time-domain cumulative bias, frequency-domain characteristics, and residual autocorrelation. The problem data is obtained by jointly analyzing sensitive parameters in the time and frequency domains. The joint analysis includes: using cumulative sums and control charts to identify the systemic lifetime prediction model offset by the cumulative deviation in the time domain; using windowed piecewise average power spectral density estimation to identify unmodeled periodic perturbations in the frequency domain; and using residual autocorrelation to identify dynamic response delay. Simulate the problem data to obtain simulated problem data, and calculate the first data: ； The first data of time t is The number of problem data is The time constant is D, and the regularization coefficient is The number of sensitive parameters is The vth sensitive parameter is The deviation of the sensitive parameter from the prior value is The data for the w-th simulation problem are The data for the wth question is Problem data The time of appearance is The time is t.

6. The product reliability assessment method based on simulation according to claim 1, characterized in that, A method for obtaining second data by comparing and analyzing aging parameters of the simulated data and the monitoring data includes: A graph neural network is used to construct the spatial correlation information of the internal structure of the power battery. Short-time Fourier transform and wavelet transform are used to extract aging parameters from simulated data and monitoring data at multiple scales. The multiple scales include macroscopic, microscopic and mesoscopic. The aging features include capacity decay rate, relaxation time distribution and internal resistance increment slope. The aging parameters include simulated aging parameters and actual aging parameters. The aging parameters are spatially mapped using a kernel function, expressed as follows: ； The x-axis of the aging parameter is d, the y-axis is g, and the coordinates of the aging parameter are... The kernel function is The standard deviation of aging parameters is The adjustment parameter is The cycle period is ; A deep feature matching mechanism is introduced to measure aging similarity, and aging similarity is calculated based on dynamic weights: ； ； The aging similarity at time t is The dynamic weight of the r-th aging parameter at time t is The dynamic weight of the r-th aging parameter at time t+1 is: The r-th simulated aging parameter is The r-th actual aging parameter is The learning rate is The gradient of the r-th aging similarity is The gradient of the p-th aging similarity is The dynamic weight of the p-th aging parameter is The number of aging parameters is ; Simulated data and monitoring data are paired based on aging similarity to obtain paired data, and the standard deviation of the paired data is used as the second data.

7. The product reliability assessment method based on simulation according to claim 1, characterized in that, A method for constructing a power battery reliability assessment model based on the first data and the second data includes: The actual remaining battery life is obtained based on monitoring data. An objective function is constructed based on the first and second data points, and its expression is: ； Where the objective function for the j-th evaluation at time t is: The simulated remaining service life is The actual remaining service life is The loss functions for simulated remaining useful life and actual remaining useful life are as follows: The first data of the j-th evaluation at time t is The second data from the j-th evaluation at time t is ; The reliability assessment model for power batteries includes self-encoders, long short-term memory networks, and artificial neural network algorithms; An autoencoder encodes input data into a low-dimensional representation, decodes and restores it, learns the inherent characteristics of the input data, and extracts key features based on their correlation with product reliability. Long Short-Term Memory (LSTM) networks selectively remember and forget key long-term and short-term information through memory units and gating mechanisms, and obtain temporal characteristics by performing temporal characteristic analysis on long-term and short-term information. Artificial neural network algorithms are based on objective functions. Through network training and learning, they can uncover the potential relationship between time-series characteristics and the reliability of power battery products, and obtain reliability assessment values ​​based on the potential relationship.

8. A product reliability assessment system based on simulation, used to perform the method according to any one of claims 1-7, characterized in that, include: Data acquisition module: used to collect monitoring data and related data of preset products, and to preprocess the monitoring data and related data; The relevant data includes environmental data, historical operating data, manufacturing process data, and design parameters of the power battery; the monitoring data includes historical data and data to be evaluated; the historical data includes electrical parameters, control signals, thermal parameters, and mechanical parameters. The simulation module is used to obtain simulation requirement data by fusing and inverting the relevant data and the monitoring data, using the environmental data as the simulation background, and performing accelerated simulation of the power battery system based on the simulation requirement data and the simulation background to obtain simulation data. The simulation requirement data includes electrode material characteristics, separator parameters, single cell data, module data, electrothermal coupling data, electromechanical coupling data, operating condition and environmental data, chemical degradation parameters, mechanical failure data, BMS strategy parameters, fault response logic, in-situ test data, and electrochemical test data. Prediction Deviation Analysis Module: Used to predict the lifespan of the simulated data to obtain simulated lifespan data, monitor the lifespan based on the data to be analyzed to obtain actual lifespan data, and perform deviation analysis based on the simulated lifespan data and the actual lifespan data to obtain first data; Modeling and evaluation module: used to obtain second data by comparing and analyzing aging parameters of the simulation data and the monitoring data, construct a power battery reliability evaluation model based on the first data and the second data, input the data to be evaluated into the power battery reliability evaluation model, and output the evaluation results.

Images