A product reliability prediction method, device, medium and product in a multi-stage scenario

Abstract

The application discloses a product reliability prediction method and device in a multi-stage scenario, a medium and a product, relates to the field of product reliability evaluation, and comprises the following steps: acquiring original environment data and corresponding reliability data of a product in each stage; determining Weibull parameters of a corresponding stage based on a multi-stage equivalent time model and a group algorithm according to the reliability data of each stage; determining a Weibull parameter prediction model by using a random forest model according to the original environment data of each stage and the Weibull parameters of the corresponding stage; and performing reliability prediction by using the trained Weibull parameter prediction model according to original environment data of a new stage of the product.

Description

Technical Field

[0001] This application relates to the field of product reliability assessment technology, and in particular to a product reliability prediction method, device, medium and product in a multi-stage scenario. Background Technology

[0002] From the moment a product leaves the factory, it faces different environmental stages, each with varying environmental stresses. These different environmental stresses determine the changes in product reliability. For example, a product stored in a temperature- and humidity-controlled cave after leaving the factory can have a lifespan of up to 10 years. After 10 years, the product's reliability still meets expectations. However, products from the same batch that are immediately deployed at sea after leaving the factory, experiencing harsh environments such as high and low temperature shocks and salt spray corrosion, will have their lifespan drastically shortened.

[0003] To determine a product's lifespan under different environments, current research typically employs statistical methods to fit its reliability distribution (e.g., the Weibull distribution). However, previous studies have only analyzed single environmental conditions and their classification of environmental factors is rather coarse. For example, a manufacturer might state that a product's reliability is 10 years in a 25°C constant-temperature warehouse environment, but only 100 days in harsh environments. This method neither covers nor can predict all possible future environmental conditions, making it difficult to accurately calculate the product's actual lifespan in unknown environments.

[0004] Currently, the common methods for determining product reliability mainly involve using accelerated aging tests and product lifespan statistics to determine whether the product conforms to a certain single-stage reliability model (such as the Weibull distribution), and then using this model to estimate its lifespan. A typical Weibull distribution is as follows: ;in, The cumulative failure probability, For time, For scale parameters, m The shape parameter is used to estimate the lifetime using the Weibull distribution formula. It is evident that the key to estimating lifetime using this formula lies in determining the shape and scale parameters. The shape parameter determines the shape of the distribution and the trend of the failure rate over time. When the failure mode remains constant, the shape parameter does not change significantly under different environmental stress levels. For example, high temperature-induced brittle fracture failure results in a larger shape parameter, while humidity-induced corrosion fracture failure results in a smaller shape parameter. The scale parameter represents the characteristic lifetime; a larger scale parameter indicates a longer product lifespan. The scale parameter decreases with increasing environmental stress, meaning high stress accelerates failure. Traditional methods for estimating single-stage Weibull parameters include gradient descent and maximum likelihood estimation. However, when the number of stages is expanded from single-stage to multi-stage, it is prone to getting trapped in local optima in the high-dimensional parameter space, leading to significant prediction bias.

[0005] Furthermore, the derivation of multi-stage Weibull parameters must consider the issues of failure probability superposition and estimation. Assuming a product is always in a single environment (e.g., constant temperature storage) and its aging pattern is fully known (e.g., conforming to a Weibull distribution), its reliability changes can be directly estimated using formulas, thus estimating its remaining lifespan (where lifespan refers to the time it takes for the failure probability to rise to a certain point; for example, if the failure probability of a product rises to 85% in 10 years, then its lifespan is considered to be 10 years). However, in reality, products often experience multiple different environmental stages, and the environmental stresses (such as temperature and humidity) in each stage accelerate product aging in different ways. Therefore, directly using formulas for estimation deviates significantly from reality. In summary, current models struggle to integrate multi-source environmental data (such as temperature and humidity fluctuations) in real time and cannot support extended predictions for new environmental stages.

[0006] Therefore, existing technologies still have the following problems: First, there is the problem of high-dimensional parameter optimization. The dimensionality of multi-stage Weibull parameters increases with the number of stages, and traditional gradient descent or maximum likelihood estimation is prone to getting trapped in local optima, leading to significant prediction bias. Second, there is the problem of the separation between physical models and data-driven models. Traditional methods rely solely on physical models or purely data-driven models; the former lacks generalization ability, and the latter lacks interpretability. Third, there is the problem of insufficient dynamic expansion capability. Traditional methods need to refit global parameters to cope with new environmental stages and cannot expand predictions in real time.

[0007] Based on the above problems, there is an urgent need to provide a new product reliability prediction method in multi-stage scenarios, which can achieve high-precision, high-efficiency, and scalable reliability product prediction. Summary of the Invention

[0008] The purpose of this application is to provide a product reliability prediction method, device, medium, and product under multi-stage scenarios, which can solve the problems of low accuracy, poor efficiency of parameter optimization, and insufficient dynamic expansion capability in product reliability prediction under multi-stage stress environment, and achieve high-precision, high-efficiency, and scalable product reliability prediction.

[0009] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a product reliability prediction method for a multi-stage scenario, the product reliability prediction method for a multi-stage scenario including: Acquire the raw environmental data and corresponding reliability data of the product at each stage; the raw environmental data includes temperature data and humidity data; the reliability data is the cumulative failure probability of each observation point with timestamps; Based on the reliability data of each stage, the Weibull parameters of the corresponding stage are determined using a multi-stage equivalent time model and an order group algorithm. Based on the original environmental data of each stage and the corresponding Weibull parameters, a random forest model is used to determine the Weibull parameter prediction model. Based on the original environmental data during the product addition phase, reliability prediction is performed using a trained Weibull parameter prediction model.

[0010] Optionally, the process of acquiring the product's raw environmental data and corresponding reliability data at each stage further includes: Using formula The original environmental data is preprocessed to obtain preprocessed original environmental data; in, For environmental feature vectors, The first environmental feature vector One portion, , For stage numbering, This represents the total number of stages. , The average daily temperature For temperature standard deviation, The highest temperature of the day, This is the lowest temperature of the day. This is the 25th percentile for temperature. This is the 75th percentile of temperature. The average daily humidity The standard deviation of humidity. The highest humidity of the day. The lowest daily humidity. The 25% quantile for humidity. This is the 75% quantile for humidity.

[0011] Optionally, determining the Weibull parameters for the corresponding stage based on the reliability data of each stage, using a multi-stage equivalent time model and an order group algorithm, specifically includes: The equivalent time length of each observation point in each stage is determined using a multi-stage equivalent time model; Based on the equivalent time length, determine the corresponding predicted cumulative failure probability; Based on the reliability data and the corresponding predicted cumulative failure probability for each stage, an objective function is constructed for each stage; the objective function is used to determine the error between the reliability data and the corresponding predicted cumulative failure probability for each stage. Based on the objective function of each stage, the Weibull parameters for the corresponding stage are determined using the order group algorithm.

[0012] Optionally, determining the equivalent time length of each observation point in each stage using a multi-stage equivalent time model specifically includes: Using formula Determine the equivalent time length of each observation point in each stage; in, For the first k Phase 1 j The equivalent time length of each observation point For the first k Stage scale parameters, For the first k- Scale parameters for stage 1, For the first k- The equivalent time length of stage 1, For the first k The shape parameters of the stage For the first k- The shape parameters of stage 1 For time points The actual length of time before, k For stage numbering, j Number the observation points.

[0013] Optionally, determining the corresponding predicted cumulative failure probability based on the equivalent time length specifically includes: Using formula Determine the corresponding predicted cumulative failure probability; in, For the first k Phase 1 j The predicted cumulative failure probability corresponding to each observation point For the first k Phase 1 j The equivalent time length of each observation point For the first k Stage scale parameters, For the first k The shape parameters of the stage.

[0014] Optionally, the step of constructing the objective function for each stage based on the reliability data and the corresponding predicted cumulative failure probability for each stage specifically includes: Using formula Determine the objective function ; in, For stage numbering, This represents the total number of stages. For the first k The Weibull parameters before the stage, i.e., the first stage k All parameter combinations before the stage, For the first k The total number of observation points in the phase, j Number the observation points. For the firstk Phase 1 j The predicted cumulative failure probability corresponding to each observation point For the first k Phase 1 j The cumulative failure probability corresponding to each observation point.

[0015] Optionally, the step of using a trained Weibull parameter prediction model to perform reliability prediction based on the original environmental data of the product addition stage specifically includes: Based on the original environmental data during the product addition phase, the corresponding predicted Weibull parameters are determined using the trained Weibull parameter prediction model. Determine the corresponding predicted cumulative failure probability based on the predicted Weibull parameters; The lifespan of a product is determined by predicting the cumulative failure probability.

[0016] In a second aspect, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the product reliability prediction method in the multi-stage scenario described in any one of the above.

[0017] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the product reliability prediction method in the multi-stage scenario described above.

[0018] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the product reliability prediction method in the multi-stage scenario described above.

[0019] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a product reliability prediction method, device, medium, and product under multi-stage scenarios. By combining the original environmental data and the corresponding reliability data, and using a multi-stage equivalent time model to quantify the superposition effect of variable stress, it overcomes the limitations of traditional methods that only address single environmental conditions. It combines the order group algorithm to determine the Weibull parameters corresponding to each stage, and uses a random forest model to construct a Weibull parameter prediction model to achieve high-precision prediction of product reliability. In other words, this application supports real-time reliability assessment in complex environments and is applicable to life assessment and health management in various fields such as components, new energy batteries, and industrial equipment, providing strong technical support for product health management. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating a product reliability prediction method in a multi-stage scenario according to an embodiment of this application. Figure 2 This is a flowchart illustrating a product reliability prediction method in a multi-stage scenario according to an embodiment of this application. Figure 3 This is a schematic diagram of the Weibull parameter determination process in one embodiment of this application. Figure 3 Part (a) represents the Weibull parameters obtained by iterating the order group algorithm twice. Figure 3 Part (b) represents the Weibull parameters obtained after 14 iterations of the order group algorithm. Figure 3 Part (c) represents the Weibull parameters obtained after 16 iterations of the order group algorithm. Figure 3 (d) represents the Weibull parameters obtained after 51 iterations of the order group algorithm. Figure 4 This is a schematic diagram illustrating the convergence of the objective function in one embodiment of this application; Figure 5 This is a schematic diagram comparing the Weibull parameters and the predicted Weibull parameters in one embodiment of this application. Figure 5 Part (a) is a comparison chart of scale parameters. Figure 5 (Part (b) is a comparison chart of shape parameters). Figure 6 This is a diagram showing the number distribution of trees in the predicted Weibull parameter distribution of one embodiment of this application. Figure 6 Part (a) is a scale parameter distribution - a tree number distribution plot. Figure 6 Part (b) is a shape parameter distribution - tree number distribution diagram. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] In one exemplary embodiment, such as Figure 1 and Figure 2 As shown, a product reliability prediction method for a multi-stage scenario is provided, including the following S1 to S4. Wherein: S1: Obtain the raw environmental data and corresponding reliability data of the product at each stage.

[0025] During product use, the raw environmental data is statistically analyzed and recorded daily, and the product's performance indicators are tested regularly.

[0026] The raw environmental data includes temperature and humidity data; the purpose of testing the product's performance indicators is to determine the product's failure status and to determine the reliability data of the product at each stage. The reliability data is the cumulative failure probability of each observation point with timestamps.

[0027] Specifically, temperature and humidity data are collected in real time by high-precision sensors or can be recorded manually. Different stages are determined based on significant changes in the environment, such as storage conditions and extreme environments, to ensure that the internal environmental characteristics of each stage are relatively consistent.

[0028] Product performance testing includes monthly power-on checks, semi-annual testing using model-specific testing equipment, inspection of product fracture after each harsh environment exposure, and periodic ultrasonic testing with propellant cartridges. When product performance or condition exceeds permissible thresholds, it is marked as a failure and the number of failed products is incremented by 1. The cumulative failure probability is obtained by statistically analyzing the proportion of failed samples at each testing time point to the total sample (for example, if 3 / 50 samples fail, the cumulative failure probability is 6%).

[0029] Convert the duration of each stage into days. For example, if the product is stored outdoors from day 50 to day 100, and before outdoor storage, the product underwent two different environmental stages: warehouse storage and harsh environment, then... =50, i.e., the third stage It lasts for 50 days.

[0030] Based on the cumulative failure probability at different observation points in each stage, determine the cumulative failure probability of each observation point with a timestamp. .in, , For stage numbering, Let k be the total number of stages, where k is the variable representing the stages, and K is the upper limit of k. For the first k The total number of observation points in the phase, that is, in the first phase k During this phase, all products are tested and the number of failures is recorded. jNumber the observation points.

[0031] Based on the above stages Taking a 50-day period as an example, due to high time and testing costs, the third phase did not involve daily checks to determine product failure and thus statistical analysis. Instead, checks were only conducted on days 53, 70, and 99. 3, j= 1, 2, 3. The cumulative failure probability of the three observation points can be obtained through the above steps. , and In this application, , , =99.

[0032] Since objective conditions may differ at different stages, and recording methods, data formats, and data densities may vary, it is necessary to preprocess the raw environmental data to transform it into a standardized format suitable for algorithm training. The transformation formula is as follows: .

[0033] in, For environmental feature vectors, The first environmental feature vector One portion, , For stage numbering, This represents the total number of stages. , The average daily temperature The standard deviation of temperature reflects temperature fluctuations. The highest temperature of the day, The lowest temperature of the day. The 25th percentile for temperature reflects periods of low temperature, meaning that 25% of observed values ​​are lower than this value, or that there is a 25% probability that the observed temperature during this period will be lower than this value. The 75th percentile for temperature reflects the boundary of high-temperature periods, indicating that 75% of observed values ​​are greater than this value, or that there is a 75% probability that the observed temperature during this period is greater than this value. The average daily humidity The standard deviation of humidity. The highest humidity of the day. The lowest daily humidity. The 25% quantile for humidity. This is the 75% quantile for humidity.

[0034] In this application, the dimension c of the environmental feature vector is 12.

[0035] S2: Based on the reliability data of each stage, determine the Weibull parameters for the corresponding stage using a multi-stage equivalent time model and order group algorithm.

[0036] S2 specifically includes: S21: Use a multi-stage equivalent time model to determine the equivalent time length of each observation point in each stage.

[0037] The key assumption of the multi-stage equivalent time model recursion is that the connection between stages depends only on the cumulative failure probability at the end of the previous stage, rather than all historical data.

[0038] This method of time transfer avoids the problem of tracking all historical data, reduces computational complexity, and only requires attention to the original environmental data and corresponding reliability data of the current stage and the cumulative failure probability of the previous stage, without needing to care about the original environmental data and corresponding reliability data of previous stages.

[0039] S21 specifically includes: S211: Determine the initial conditions for the order group algorithm.

[0040] The initial Weibull parameters for each stage are randomly generated using the order group algorithm, and the parameters for the first stage are determined. k Phase 1 j The first observation point before s The actual duration of the phase.

[0041] The first in this application s The stage is the first k Any stage before the stage, i.e. It is assumed that the cumulative failure rate of the two stages at each stage transition point is equal. s The formula for the actual duration of a stage is as follows: ; in, For the first s The actual length of time of the phase For the first k Phase 1 j Time at each observation point Let s be the end time of phase s. t is the start time of phase s.

[0042] Using functions The purpose of taking the minimum value is to distinguish between the s-th stage and the t-th stage. k The positional relationship of the stages.

[0043] The formula for the complete duration of phase s is as follows: ; .

[0044] S212: Determine the equivalent time length of each observation point in each stage.

[0045] Starting from stage 1, an equivalent time recursion is performed. At the end of stage 1, the cumulative failure probability of the observation point is determined using the Weibull distribution formula, as follows: ; in, This represents the cumulative failure probability at the last observation point of Phase 1. This refers to the time of the last observation point in Phase 1. These are the scale parameters for stage 1. These are the shape parameters for the first stage.

[0046] Since the cumulative failure probability is equal at the junction of stage 1 and stage 2, the Weibull distribution curve of stage 2 is the initial value. Scale parameters of the second stage and the shape parameters of the second stage The known curve.

[0047] At the start of stage 2, if stage 1 does not exist, the cumulative failure probability of the observation points at the end of stage 1 is entirely determined by the scale parameter of stage 2. and the shape parameters of the second stage Starting from time zero, the time when the cumulative failure probability of the observation point at the end of the first stage is generated is called the "time when the first stage is equivalent to the second stage".

[0048] By inversely applying the Weibull distribution formula, we can deduce that the time for the first stage is equivalent to the time for the second stage. The calculation formula is as follows: ; in, The scale parameters for stage 2. The shape parameters for the second stage, For the first k The Weibull parameters before the stage, i.e., the first stage k All Weibull parameters prior to the stage, including scale and shape parameters.

[0049] At the end of stage 2, the cumulative failure probability generated is determined entirely by the scaling parameters of stage 2. and the shape parameters of the second stage The time length caused by this alone is called the equivalent time length of the second stage, and the calculation formula is as follows: ; in, This is the equivalent time length for stage 2. This refers to the time of the last observation point in the second phase.

[0050] At the start of Phase 3, the time of Phases 1-2 is equivalent to that of Phase 3. The calculation formula is as follows: ; in, These are the scale parameters for stage 3. These are the shape parameters for the third stage.

[0051] At the end of Phase 3, the equivalent time length of Phase 3. The calculation formula is as follows: ; in, This refers to the time of the last observation point in stage 3.

[0052] And so on, Phase 1, Phase 2, ..., Phase 3 k- In stage 1, the first... k The cumulative failure probability at the start of the phase is entirely determined by the first phase. k Stage scale parameters and the k Stage shape parameters If it occurs, then the first stage, the second stage, ..., the ... k -1 stage is equivalent to the first k Phase Time The calculation formula is: ; in, For the first k Stage scale parameters, For the first k The shape parameters of the stage For the first k- Scale parameters for stage 1, For the first k- The shape parameters of stage 1 For the first k- The equivalent time length of stage 1.

[0053] Furthermore, this application is capable of calculating the equivalent time length at any observation point within any stage. The calculation formula is as follows: ; When we get the following formula: ; in, For the firstk Phase 1 j The equivalent time length of each observation point For the first k Time points within the phase The actual length of time before, i.e., the first k From the start of the stage to the first j The length of time between observation points.

[0054] In an exemplary embodiment, a product goes through three working phases, each lasting 30 days, 60 days, and 90 days, respectively.

[0055] Observations were conducted on days 10 and 20 of phase 1, days 10, 50, and 60 of phase 2, and days 10 and 80 of phase 3. The observation results are shown in Table 1.

[0056] Table 1. Statistical table of observation points at each stage

[0057] The time calculation formula for stage 1 being equivalent to stage 2 is as follows: .

[0058] The time calculation formula for stage 3, which is equivalent to stage 1 and stage 2, is as follows: .

[0059] S22: Determine the corresponding predicted cumulative failure probability based on the equivalent time length.

[0060] After determining the equivalent time length of each observation point in each stage, the corresponding predicted cumulative failure probability can be obtained according to the Weibull distribution formula. The calculation formula is as follows: ; in, For the first k Phase 1 j The predicted cumulative failure probability corresponding to each observation point For the first k Phase 1 j The equivalent time length of each observation point.

[0061] In one exemplary embodiment, taking Table 1 as an example, it is assumed that the initial Weibull parameters are randomly given by the order group algorithm. The calculation process for the corresponding predicted cumulative failure probability is shown in Table 2.

[0062] Table 2. Calculation process of cumulative fault probability at each observation point

[0063] S23: Construct the objective function for each stage based on the reliability data and the corresponding predicted cumulative failure probability for each stage.

[0064] To determine the error between the reliability data and the corresponding predicted cumulative failure probability at each stage, an objective function is constructed. The objective function formula is as follows: .

[0065] The objective function is to subtract the reliability data and the corresponding predicted cumulative failure probability at each stage, then calculate the absolute error, and finally apply this error to the next stage. k First, calculate the sum of the absolute errors of all observation points; then calculate the sum of the absolute errors of all observation points over all stages, and finally calculate the average.

[0066] The construction and recursive mechanism of the multi-stage equivalent time model maps real time to equivalent aging time under different environmental stresses, quantifies the nonlinear effects of multi-stage stress superposition, and overcomes the limitations of traditional methods that only address single environmental conditions. Through a recursive mechanism, it dynamically integrates the effects of multi-stage stresses, supporting reliability assessment in complex time-series environments. In the embodiment, the equivalent time length is calculated recursively and the predicted cumulative failure probability is determined, verifying the quantification capability of the multi-stage equivalent time model for multi-stage stress superposition.

[0067] S24: Based on the objective function of each stage, the Weibull parameters of the corresponding stage are determined using the order group algorithm.

[0068] The order group algorithm in this application first assigns unknowns before iterative computation. Assign a specific value, i.e. randomly generate initial Weibull parameters, and then determine the error between the reliability data and the corresponding predicted cumulative failure probability for each stage according to the above steps, i.e., the objective function for each stage.

[0069] The execution flow of the order group algorithm revolves around the hierarchical evolution mechanism. By simulating the behavioral logic between rulers and laborers, it iteratively optimizes the high-dimensional Weibull parameters.

[0070] First, randomly generate a number of individuals. The total number of individuals is... The calculation formula is: Each individual's coordinates are a randomly generated initial Weibull parameter. These individuals are then grouped: all individuals are divided into several subgroups, each subgroup having a coordinate system. rulers And coordinates are Several workers ( ,in, It represents both the ruler and the subgroup number. Number the workers within each subgroup. Specifically, the number of subgroups. Similar to the number of rulers, the number of subgroups is used to control population diversity, specifically the number of laborers in each subgroup. The local search capability is determined. The ruler stores the current optimal Weibull parameters and performs the search according to the search step size, which is specified in this application. The worker is responsible for performing specific search tasks according to the search step size mechanism. The local search decay coefficient controls the step size contraction speed when the worker approaches the ruler. The local search decay coefficient in this application... This division of labor ensures a large search range to avoid getting trapped in local optima, while also allowing for refined searches near advantageous solutions to guarantee search accuracy, thus balancing the contradiction between solution space exploration and local refinement.

[0071] After the loop begins, the worker roaming phase starts: all workers adaptively move with the ruler as the reference point—exploring new areas with large strides when moving away from the ruler, and exponentially shrinking strides to achieve refined development when moving closer; if a worker finds a solution superior to the ruler's (evaluated by the objective function), a role-swapping mechanism is immediately triggered, allowing the ruler to claim the achievement, ensuring continuous optimization of the search direction. Specifically, the worker position update formula is as follows: ; ; ; in, For the updated worker coordinates, The Euclidean distance between the individual and the ruler is used to control the decay of the search step size. To be Cut off to interval middle, To minimize the initial Weibull parameters, To maximize the initial Weibull parameter, .

[0072] Role swapping mechanism, if it exists satisfy If the positions of the laborer and the ruler are to be exchanged, the exchange formula is as follows: ;in, This involves swapping the positions of individuals.

[0073] After each preset number of worker walks iterations, a hierarchy replacement phase is initiated: by comparing the fitness values ​​of the rulers of all subgroups (the optimal solution for each subgroup), the worst-performing subgroup (including one ruler and all workers within that subgroup) is eliminated, and its computational resources are dynamically allocated to the remaining subgroups, with the globally optimal subgroup receiving an additional resource quota. Specifically, the elimination process is as follows: Next, one group remains at the end. In all... Among them, the worst will be eliminated. , The dominant entity that maximizes the objective function. The number is assigned, and all individuals in the subgroup with the worst objective function are removed. （ A new subgroup is generated and resources are reallocated. The allocation method is as follows: the remaining groups share the resources equally, and if the distribution is not evenly divided, the remainder is allocated to the current best group. This forces the population to focus on the dominant parameter region and prevents premature convergence.

[0074] The above "walk-and-replace" loop is executed repeatedly until the convergence condition is met or the maximum number of iterations is reached. (This process is repeated several times), and the final output of the globally optimal ruler is the desired result, yielding the optimal Weibull parameters, which are the Weibull parameters for the corresponding stage.

[0075] The order group algorithm avoids the blind search in high-dimensional space by using hierarchical division of labor, overcomes the local optimum trap by resource reorganization, and achieves incremental optimization from coarse screening to fine refinement by using distance correlation step size, providing high-precision input for subsequent random forest models and laying the foundation for multi-stage reliability prediction. Addressing the challenges of unpredictable computational efficiency due to dynamic changes in the objective function's parameter size and the high computational complexity of the time recursion process, this application's order group algorithm avoids invalid computation by locking the iteration upper limit mechanism, optimizes computational power allocation by combining a dynamic resource recycling mechanism, and adopts a minimalist computational architecture (retaining only dynamic step size adjustment) to minimize computational overhead. This significantly improves engineering interpretability and computational efficiency while ensuring the mathematical optimality of the output results. Furthermore, the dynamic step size order group algorithm optimizes high-dimensional Weibull parameters, overcoming the problem of traditional gradient descent and maximum likelihood estimation easily getting trapped in local optima in high-dimensional parameter spaces. It balances computational efficiency and accuracy through dynamic step size and population reorganization. This application proposes an adaptive step size adjustment mechanism combined with a population reorganization strategy (eliminating the worst subgroup and allocating remaining resources), which can improve the global search efficiency of multi-stage Weibull parameters.

[0076] S3: Based on the original environmental data of each stage and the corresponding Weibull parameters, a random forest model is used to determine the Weibull parameter prediction model.

[0077] S3 specifically includes: S31: Data preparation.

[0078] The components of each stage are extracted from the preprocessed raw environmental data, i.e., the environmental feature vector. As an independent sample, a total One sample, Each sample is used as input. The Weibull parameters for the corresponding stage, calculated by the order group algorithm, are used as output; for example, the output for stage 1 is the scale parameter for stage 1. and the shape parameters of stage 1 The output of the second stage is the scale parameter of the second stage. and the shape parameters of the second stage .

[0079] S32: Construct the training set.

[0080] right Each sample was randomly drawn with replacement. Next, the Weibull parameters from each stage are used together as a training set. The training set expression is as follows: .

[0081] In one exemplary embodiment, there are three stages, where stage 1 is selected twice and stage 2 is not selected. The above sampling is performed independently B times to obtain B training sets.

[0082] S33: Train on a single tree.

[0083] Training prediction scale parameters The Taking the tree as an example, the input is the first tree. training set The output is the first... Trees. Repeat B You can get it in one go. B a tree, then this B The tree is the prediction scale parameter. Predict shape parameters The determination process is the same as described above. Prediction scale parameters and predicted shape parameters These are collectively referred to as the predicted Weibull parameters.

[0084] Specifically, the root node is initialized first, and the node contains all samples. , This is the sample for the current node. If the current node is the root node, (Because auto-sampling may contain duplicate samples). Set the current root node depth. .

[0085] Then, a cyclical split is performed, first checking if the training process meets the termination condition. If it does... ( To preset the minimum number of samples, we can choose 5). ( To preset the maximum depth, a value of 10 can be used, along with the scale parameters of samples within the node. The variance of the value is less than the threshold. The loop splitting stops when any of the conditions (0.1) are met.

[0086] Candidate features are randomly selected; specifically, they are randomly drawn without replacement from the dimensions of the 12 environmental feature vectors. There are 10 features, which are the candidate features. Let be the floor function, representing rounding down. Then, iterate through the candidate features and the split threshold; specifically, for each candidate feature (e.g., ... )extract The value of this feature in all samples The values ​​of this feature are sorted and deduplicated, and then the midpoint between adjacent values ​​is calculated as a candidate threshold. The sorting formula and the formula for calculating adjacent center points are as follows: ; ; in, The unique value sequence obtained after sorting and deduplication is the first one. v One value, This represents the number of feature values ​​after sorting and deduplication.

[0087] In one exemplary embodiment, assume the current node There are 5 samples, and candidate features (such as...) The values ​​of are [25.0, 26.0, 25.0, 28.0, 26.0], the sorted sequence is [25.0, 25.0, 26.0, 26.0, 28.0], and the unique value sequence after deduplication is [25.0, 26.0, 28.0]. ， , , Number of eigenvalues Adjacent center points The calculation process is as follows: and .

[0088] After determining the candidate thresholds, the optimal split point is calculated. Specifically, each candidate feature is analyzed. At each of its candidate thresholds The disposal is divided into two subsets, left and right; the left subset is... satisfy The sample; the right subset satisfy The sample.

[0089] Definition Formula And calculate MSE The amount of decrease The formula for calculating the decrease is as follows: .

[0090] in, For sample set The mean square error, The subset of samples contained in the current tree node. i For samples in the set Index in For sample set The Middle The scale parameter of each sample. For sample set All The arithmetic mean.

[0091] After determining the descent value, select the optimal split; specifically, select the split that minimizes the descent value. The largest Combine. Set update depth Record and continue the above splitting process for the left and right subsets. Repeat the above steps until the termination condition is met, and record the value of each node. Combinations, in which, These are the features, i.e., the candidate features that have been determined. The threshold, i.e., the determined candidate threshold, is used to further record the leaf node values. : .

[0092] This application embeds Weibull parameters into a random forest model. The optimized Weibull parameters, i.e., the Weibull parameters of the corresponding stage, are jointly input into the random forest model along with the environmental feature vector to train a Weibull parameter prediction model. This breaks through the bottleneck of the separation between traditional physical models and data-driven models, enhances the interpretability of machine learning models through physical parameters (Weibull distribution), and utilizes the generalization ability of random forests to adapt to complex environments.

[0093] S4: Based on the original environmental data during the product addition phase, use the trained Weibull parameter prediction model to perform reliability prediction.

[0094] S4 specifically includes: S41: Based on the original environmental data of the new product phase, determine the corresponding predicted Weibull parameters using the trained Weibull parameter prediction model.

[0095] Input the original environmental data for the new product stage, and determine the stage boundaries based on the points of significant change in environmental stress. to ,in For the first K The end time of the phase is also the [missing information]. K The start time of Phase +1 For the stage K +1 is the end time, which is also the time point at which the cumulative failure probability is to be obtained. The original environmental data for the new phase is preprocessed to obtain the corresponding components of the environmental feature vector, i.e., the components of the environmental feature vector for the new phase: ; in, The first environmental feature vector of the newly added stage Each component.

[0096] Specifically, will By inputting a single tree and performing operations on the splitting process, a component of the predicted Weibull parameter can be obtained. The predicted Weibull parameter can be obtained by averaging the components of the predicted Weibull parameter obtained from each tree. This can be expressed by the formula: ; ; in, For the first Stage scale parameters, For the first The shape parameters of the stage.

[0097] S42: Determine the corresponding predicted cumulative failure probability based on the predicted Weibull parameters.

[0098] S43: Determine product lifespan by predicting cumulative failure probability.

[0099] Seamless prediction for new environmental phases: The predicted Weibull parameters for new phases are determined using a pre-trained Weibull parameter prediction model, eliminating the need for global model retraining. This solves the problem of traditional methods requiring refitting global parameters and supports real-time expansion of prediction capabilities. Example predictions validate the feasibility of dynamic expansion.

[0100] In another exemplary embodiment, a batch of 100 products is selected for study. This includes the following steps S1-S4.

[0101] S1: Obtain the raw environmental data and corresponding reliability data of the product at each stage.

[0102] Based on the environment, the system can be divided into 7 stages, and the reliability data for each stage is shown in Table 3: Table 3. Reliability data statistics for each stage

[0103] Using temperature and humidity sensors and warehouse statistics, the raw environmental data of the products were obtained and preprocessed. The preprocessed raw environmental data is shown in Table 4.

[0104] Table 4. Raw environmental data after pretreatment

[0105] S2: Based on the reliability data of each stage, determine the Weibull parameters for the corresponding stage using a multi-stage equivalent time model and order group algorithm.

[0106] Suppose the coordinates of a volume in the order group algorithm are: .

[0107] (1) Phase 1 ( k =1).

[0108] The calculation table for the first stage of observation points is shown in Table 5.

[0109] Table 5 Calculation Table for Observation Points in Phase 1

[0110] (2) Stage 2 ( k =2).

[0111] Phase 1 to Phase 2 transition: .

[0112] The calculation table for the observation points in the second stage is shown in Table 6.

[0113] Table 6 Calculation Table for Observation Points in Phase 2

[0114] (3) Stage 3 ( k =3).

[0115] Transition from Phase 1-2 to Phase 3: .

[0116] The calculation table for the observation points in the third stage is shown in Table 7.

[0117] Table 7 Calculation Table for Observation Points in Stage 3

[0118] (4) Stage 4 k =4).

[0119] Transition from Stage 1-3 to Stage 4: .

[0120] The calculation table for the observation points in the fourth stage is shown in Table 8.

[0121] Table 8 Calculation Table for Observation Points in Stage 4

[0122] (5) Stage 5 k =5).

[0123] Transition from Stages 1-4 to Stage 5: .

[0124] The calculation table for the observation points in the fifth stage is shown in Table 9.

[0125] Table 9 Calculation Table for Observation Points in Stage 5

[0126] (6) Stage 6 (k=6).

[0127] Transition from Stages 1-5 to Stage 6: .

[0128] The calculation table for the observation points in the 6th stage is shown in Table 10.

[0129] Table 10 Calculation Table for Observation Points in Stage 6

[0130] Calculate the objective function The absolute error calculation table is shown in Table 11.

[0131] Table 11 Absolute Error Calculation Table

[0132] Objective function value: Clearly, this individual error is relatively large and is not the optimal solution.

[0133] Set the parameters for the order group algorithm: (number of subgroups), (Number of workers in the subgroup) , After 51 iterations, the algorithm outputs the Weibull parameters for the corresponding stage: Phase 1: μ= 1406.94, m= 1.80; Phase 2: μ= 1915.15, m= 4.80; Phase 3: μ= 572.96, m= 4.09; Phase 4: μ= 590.74, m= 3.83; Phase 5: μ= 824.88, m= 2.28; Phase 6: μ= 2737.39, m= 1.90.

[0134] Screenshots of some iteration processes and the optimization trend of the objective function are shown below. Figure 3 As shown (only a few are displayed due to dimensional limitations) The first two dimensions, namely the scale parameter of stage 1 and the shape parameter of stage 1). Figure 3 A schematic diagram of the convergence of the objective function is shown below. Figure 4 As shown.

[0135] S3: Based on the original environmental data of each stage and the corresponding Weibull parameters, a random forest model is used to determine the Weibull parameter prediction model.

[0136] The input to the random forest model is an environmental feature vector. (6 stages × 12 dimensions), the output is In .

[0137] The random forest model is trained by setting parameters, including the number of trees. Minimum number of samples Maximum depth The trained random forest model is the Weibull parameter prediction model, which contains 100 trees. When a new environment vector is input into this tree, its corresponding scale and shape parameters can be predicted. The average of the results from the 100 trees yields the predicted Weibull parameters for the new stage. A comparison of the predicted Weibull parameters determined by the Weibull parameter prediction model and the Weibull parameters fitted by the order group algorithm is shown in the figure. Figure 5 As shown.

[0138] S4: Based on the original environmental data during the product addition phase, use the trained Weibull parameter prediction model to perform reliability prediction.

[0139] The product has entered a new stage (stage 7), and its temperature and humidity data have been processed through feature extraction as follows: .

[0140] Call the Weibull parameter prediction model trained in the first 6 stages, and Input, and obtain the predicted Weibull parameters for stage 7, including the histogram of the predicted distribution of each tree and the mean prediction results as shown below. Figure 6 As shown, the predicted Weibull parameters for stage 7 are: .

[0141] To obtain the cumulative failure probability of the seventh stage, the equivalent time length of each stage is calculated step by step, up to the end of the sixth stage. Simultaneously, to verify the Weibull parameters... The effect is to reduce the cumulative failure probability at each stage. Calculate them together, and use them in conjunction with Table 1. Comparison: (1) Phase 1.

[0142] Observation point: .

[0143] .

[0144] .

[0145] .

[0146] .

[0147] .

[0148] .

[0149] Equivalent time length at the end of Phase 1: .

[0150] Phase 1 to Phase 2 transition.

[0151] .

[0152] (2) Stage 2.

[0153] Observation point: .

[0154] .

[0155] .

[0156] .

[0157] .

[0158] .

[0159] .

[0160] Equivalent time length at the end of Phase 2: .

[0161] Phase (1, 2) → Phase 3 transition.

[0162] .

[0163] (3) Stage 3.

[0164] Observation point: .

[0165] .

[0166] .

[0167] .

[0168] .

[0169] .

[0170] .

[0171] Equivalent time length at the end of Phase 3: .

[0172] Phase (1-3) → Phase 4 transition.

[0173] .

[0174] (4) Stage 4.

[0175] Observation point ： .

[0176] .

[0177] .

[0178] .

[0179] .

[0180] .

[0181] .

[0182] Equivalent time length at the end of Phase 4: .

[0183] Phase (1-4) → Phase 5 transition.

[0184] .

[0185] (5) Stage 5.

[0186] Observation point: .

[0187] .

[0188] .

[0189] .

[0190] .

[0191] .

[0192] .

[0193] Equivalent time length at the end of stage 5: .

[0194] Phase (1-5) → Phase 6 transition.

[0195] .

[0196] (6) Stage 6.

[0197] Observation point: .

[0198] .

[0199] .

[0200] .

[0201] Equivalent time length at the end of stage 6: .

[0202] Phase 7 Reliability Calculation.

[0203] Historical equivalent time conversion:

[0204] .

[0205] The total equivalent duration of the new phase: .

[0206] Cumulative failure probability prediction:

[0207] .

[0208] Result: On the 100th day after the end of prediction phase 7, the cumulative failure probability was 37.9%.

[0209] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal. The computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is connected to the system bus via the I / O interfaces. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores product reliability prediction data under multi-stage scenarios. The I / O interfaces of the computer device are used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a product reliability prediction method under multi-stage scenarios.

[0210] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0211] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0212] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0213] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0214] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0215] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0216] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0217] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A product reliability prediction method in a multi-stage scenario, characterized in that, The product reliability prediction method under the multi-stage scenario includes: Acquire the raw environmental data and corresponding reliability data of the product at each stage; the raw environmental data includes temperature data and humidity data; the reliability data is the cumulative failure probability of each observation point with timestamps; Based on the reliability data of each stage, the Weibull parameters of the corresponding stage are determined using a multi-stage equivalent time model and an order group algorithm. Based on the original environmental data of each stage and the corresponding Weibull parameters, a random forest model is used to determine the Weibull parameter prediction model. Based on the original environmental data during the product addition phase, reliability prediction is performed using a trained Weibull parameter prediction model.

2. The product reliability prediction method under multi-stage scenarios according to claim 1, characterized in that, The process of acquiring the product's raw environmental data and corresponding reliability data at each stage further includes: Using formula The original environmental data is preprocessed to obtain preprocessed original environmental data; in, For environmental feature vectors, The first environmental feature vector One portion, , For stage numbering, This represents the total number of stages. , The average daily temperature For temperature standard deviation, The highest temperature of the day, The lowest temperature of the day. This is the 25th percentile for temperature. This is the 75th percentile of temperature. The average daily humidity The standard deviation of humidity. The highest humidity of the day. The lowest daily humidity. The 25% quantile for humidity. This is the 75% quantile for humidity.

3. The product reliability prediction method under multi-stage scenarios according to claim 1, characterized in that, The process of determining the Weibull parameters for each stage based on the reliability data of each stage, using a multi-stage equivalent time model and an order group algorithm, specifically includes: The equivalent time length of each observation point in each stage is determined using a multi-stage equivalent time model; Based on the equivalent time length, determine the corresponding predicted cumulative failure probability; Based on the reliability data and the corresponding predicted cumulative failure probability for each stage, an objective function is constructed for each stage; the objective function is used to determine the error between the reliability data and the corresponding predicted cumulative failure probability for each stage. Based on the objective function of each stage, the Weibull parameters for the corresponding stage are determined using the order group algorithm.

4. The product reliability prediction method in a multi-stage scenario according to claim 3, characterized in that, The determination of the equivalent time length of each observation point in each stage using a multi-stage equivalent time model specifically includes: Using formula Determine the equivalent time length of each observation point in each stage; in, For the first k Phase 1 j The equivalent time length of each observation point For the first k Stage scale parameters, For the first k- Scale parameters for stage 1, For the first k- The equivalent time length of stage 1, For the first k The shape parameters of the stage For the first k- The shape parameters of stage 1 For time points The actual length of time before, k For stage numbering, j Number the observation points.

5. The product reliability prediction method in a multi-stage scenario according to claim 3, characterized in that, The step of determining the corresponding predicted cumulative failure probability based on the equivalent time length specifically includes: Using formula Determine the corresponding predicted cumulative failure probability; in, For the first k Phase 1 j The predicted cumulative failure probability corresponding to each observation point For the first k Phase 1 j The equivalent time length of each observation point For the first k Stage scale parameters, For the first k The shape parameters of the stage.

6. The product reliability prediction method in a multi-stage scenario according to claim 3, characterized in that, The objective function for each stage is constructed based on the reliability data and the corresponding predicted cumulative failure probability, specifically including: Using formula Determine the objective function ; in, For stage numbering, This represents the total number of stages. For the first k The Weibull parameters before the stage, i.e., the first stage k All parameter combinations before the stage, For the first k The total number of observation points in the phase, j Number the observation points. For the first k Phase 1 j The predicted cumulative failure probability corresponding to each observation point For the first k Phase 1 j The cumulative failure probability corresponding to each observation point.

7. The product reliability prediction method under multi-stage scenarios according to claim 1, characterized in that, The process of making reliability predictions based on the original environmental data from the new product launch phase, using a trained Weibull parameter prediction model, specifically includes: Based on the original environmental data during the product addition phase, the corresponding predicted Weibull parameters are determined using the trained Weibull parameter prediction model. Determine the corresponding predicted cumulative failure probability based on the predicted Weibull parameters; The lifespan of a product is determined by predicting the cumulative failure probability.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the product reliability prediction method in a multi-stage scenario as described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the product reliability prediction method under a multi-stage scenario as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the product reliability prediction method under a multi-stage scenario as described in any one of claims 1-7.

Images