A reliability detection method for real-time mixed-line production manufacturing line

Abstract

The application relates to a reliability detection method for a real-time mixed-line production manufacturing line, which comprises the following steps: S1, acquiring historical working conditions and fault data of the mixed-line production manufacturing line, analyzing and counting key failure factors, and then hierarchically constructing a bottom-layer equipment subsystem failure Bayesian subnetwork and an upper-layer whole-line process propagation Bayesian main network; S2, dynamically and dynamically updating equipment subsystem and whole-line failure probabilities in real time based on the probability inference ability of the double-layer Bayesian network; and S3, evaluating the reliability of the mixed-line production manufacturing line according to the equipment subsystem and whole-line failure probabilities. The application can effectively reduce the fault mis-detection and missed-detection rates of the mixed-line production manufacturing line, reduce non-scheduled shutdown loss and batch production quality defects of the line, and significantly improve the operation reliability, fault treatment precision and intelligent operation and maintenance automation level of the mixed-line production manufacturing line.

Description

Technical Field

[0001] This invention relates to the field of reliability testing technology, and in particular to a reliability testing method for real-time mixed-line production lines. Background Technology

[0002] Against the backdrop of intelligent manufacturing transformation and upgrading and the widespread adoption of flexible mixed-line production models, real-time mixed-line production, characterized by multi-variety, small-batch production, has become a core production method for discrete manufacturing to improve quality and efficiency and adapt to personalized production needs. It is widely used in industrial fields such as precision machinery, automotive parts, and high-end equipment. Mixed-line production lines integrate multiple heterogeneous subsystems, including electrical, CNC, hydraulic, cooling, spindle, and feed systems. Each piece of equipment and process is deeply coupled and operates in a coordinated manner, exhibiting typical characteristics such as dynamic and changing operating conditions, complex process links, and frequent equipment interactions. As the automation and integration levels of production lines continue to increase and production cycles accelerate, real-time reliability monitoring and risk prediction of production line operation have become crucial for ensuring the continuity of mixed-line production, avoiding quality defects in mass production, and reducing equipment downtime losses. In real-time mixed-line production, minor faults in various subsystems can easily propagate and spread along the process chain, and single-point hidden anomalies can easily evolve into problems such as line-wide shutdowns, production rhythm disruptions, and product processing errors. Traditional operation and maintenance models that rely on manual inspection and post-failure review are no longer suitable for the high-precision, high-real-time requirements of mixed-line production management. How to accurately decouple multi-process coupling interference, quantify the failure propagation patterns of the production line, identify potential fault risks in real time, and achieve equipment fault tracing and dynamic assessment of overall line reliability has become a key technical problem that urgently needs to be solved in the field of reliability testing for real-time mixed-line manufacturing.

[0003] Currently, for reliability testing and fault analysis of mixed-line manufacturing production lines, existing technologies mainly rely on fault tree analysis, Markov state assessment, single-layer Bayesian networks, and traditional machine learning diagnostic methods. These technologies have been widely applied to production line equipment condition monitoring and fault diagnosis. Most existing conventional solutions achieve the identification of explicit faults and simple state assessments of production line equipment by constructing a single fault analysis model, defining fault logic based on expert experience, or fitting fault characteristics based on operational data. While these methods can achieve basic fault alarms and state judgments in dedicated production lines with single equipment and fixed processes, improving the basic efficiency of production line operation and maintenance to some extent, they have significant technical limitations under the complex conditions of real-time mixed-line production. Furthermore, existing technologies generally adopt a single-layer modeling architecture, failing to distinguish between the two-layer logic of failure evolution within equipment subsystems and the cross-process, line-wide fault propagation. This makes it difficult to effectively decouple coupling interference factors from multiple systems and processes, resulting in low accuracy in identifying latent, early-stage potential failure factors. Meanwhile, existing solutions cannot quantify the fault propagation paths, propagation probabilities, and coupling impact strength across subsystems, making it difficult to dynamically predict the overall risk evolution trend of the production line, and resulting in insufficient accuracy in fault root cause localization. In addition, traditional models have weak generalization capabilities, cannot adapt to scenarios with frequent model changes and dynamic process adjustments in mixed-line production, and their real-time inference efficiency is insufficient to meet online detection requirements. Furthermore, they generally lack hierarchical and refined reliability quantitative assessment capabilities, failing to provide comprehensive and accurate quantitative decision support for early warning, preventive maintenance, and production scheduling optimization in mixed-line manufacturing production lines.

[0004] Existing reliability testing methods for mixed-line manufacturing production lines generally employ single-layer modeling and single-fault analysis for production line status assessment and fault diagnosis. This approach is ill-suited to the complex scenarios of real-time mixed-line production, characterized by multi-process coupling, dynamic and changing operating conditions, and chain-like fault propagation. This results in low reliability testing accuracy, weak fault tracing capabilities, and a lack of dynamic risk management, failing to meet the demands for intelligent and real-time production line operation and maintenance management. The reasons are as follows: First, existing technologies often employ flat modeling methods such as single-layer Bayesian networks, fault trees, and machine learning, failing to distinguish between the hierarchical logic of failure evolution within equipment subsystems and cross-process, line-wide fault propagation. The coupling interference between multiple devices and processes cannot be effectively decoupled, leading to the model's inability to accurately identify primary and secondary failure factors, and making it difficult to effectively identify hidden faults and early potential risks. Second, existing modeling schemes fragment the process coupling relationships between various subsystems in mixed-line manufacturing production lines, enabling independent fault analysis of single devices and systems. They cannot concretely and quantitatively characterize the cross-system propagation path, propagation probability, and coupling impact strength of faults. Furthermore, existing conventional models cannot extrapolate the dynamic evolution of faults as they spread through production processes, resulting in vague root cause localization and unpredictable risk scope. Finally, the fixed topology and poor parameter adaptability of existing conventional models mean that, in the face of frequent model changes and dynamic process adjustments in mixed-line production, the model topology and parameters need to be completely reconstructed. Their generalization ability and real-time reasoning efficiency are insufficient, making it difficult to achieve online real-time detection of production line reliability. Moreover, the lack of a hierarchical and quantitative risk assessment mechanism makes it impossible to provide accurate and comprehensive quantitative decision-making basis for pre-warning, preventive maintenance, and production scheduling optimization. This can easily lead to problems such as complete line downtime and mass production quality defects due to untimely risk prediction. Summary of the Invention

[0005] The purpose of this invention is to provide a reliability testing method for real-time mixed-line manufacturing production lines to solve the aforementioned technical problems. Therefore, the technical solution adopted by this invention is as follows: A reliability testing method for real-time mixed-line manufacturing lines includes the following steps: S1. Obtain historical operating conditions and fault data of mixed-line production lines, analyze and statistically analyze key failure factors, and then construct a layered Bayesian sub-network of failures in the underlying equipment subsystems and a Bayesian main network of process propagation in the upper-level whole line. S2. Based on the probabilistic reasoning capability of a two-layer Bayesian network, dynamically and in real-time update the failure probability of equipment subsystems and the entire production line; S3. Evaluate the reliability of mixed-line production lines based on the failure probabilities of equipment subsystems and the entire line.

[0006] In one embodiment, in S1, the historical operating conditions and fault data are a table of failure points for one year.

[0007] In one embodiment, the underlying device subsystem failure Bayesian subnetwork consists of failure root causes, failure modes, and top event nodes. The failure root causes are the contents recorded in the failure list and the source of the problem. The failure modes are the cases where the subsystem partially fails due to multiple failure root causes. The top event nodes are the subsystem failure behaviors that are directly triggered.

[0008] In one embodiment, the failure modes include mechanical failure, scheduling and communication failure, sensing and positioning failure, docking failure, task scheduling failure, communication failure, data acquisition failure, data consistency failure, and quality management failure.

[0009] In one embodiment, the instantaneous failure rate of the mechanical failure for: , in, For scale parameters; For shape parameters; The average failure rate of the scheduling communication failure for: , in, To determine the total number of faults that can ultimately be exposed, For the fault exposure rate, The duration; Instantaneous failure rate of sensing and positioning systems for: , , in, For baseline scale parameters, The baseline shape parameter is , and the covariates are . , speed, For load, Dust level, Reflective intensity, To locate the residuals, the covariate coefficients are: ; Average failure probability of docking failure for: , Among them, covariates ,in, For positioning error, For attitude deviation, For docking frequency, For fixture consistency deviation, Due to landmark identification bias, This refers to the tolerance variation during docking; for Scale parameters; For trend parameters; The duration; This is the vector of covariate coefficients.

[0010] In one embodiment, the failure probability of the top event node is calculated using a conditional probability table (CPT) based on the maximum likelihood estimation (MLE).

[0011] In one embodiment, calculating the conditional probability table (CPT) based on the maximum likelihood estimate (MLE) specifically includes: Construct the initial structure of CPT: determine the parent node of each node and the dimensions of the CPT table; Calculate the likelihood function: For each CPT, calculate the probability of the observed data occurring given the parameters. The likelihood function is the product of the joint probabilities of all observed data points. ;in, For given parameters At that time, the likelihood function of all historical observation data; This represents the total number of historical observation samples. For the first Index of each observation sample; For the first The combination of parent node states of the target node in each sample; For the first The observation status of the target node in each sample; For the parent node to be in state Under the condition that the target node is in state The conditional probability; This is the set of parameters to be estimated in the conditional probability table of the Bayesian network; To maximize the likelihood function: Adjust the parameters in CPT to maximize the likelihood function. ,in, Let be the log-likelihood function corresponding to the likelihood function. ; These are the optimal conditional probability parameters obtained through maximum likelihood estimation.

[0012] In one embodiment, the failure probability of the device subsystem is calculated using the same method as the failure probability of the top event node.

[0013] In one embodiment, in S2, the overall line failure probability is expressed as: , in, This indicates the working state of the i-th subsystem. The participation status is indicated by 0 for no participation and 1 for participation; This represents the failure probability of the i-th subsystem.

[0014] In one embodiment, S3 includes: The real-time failure probability sequence of each node is smoothed using an exponentially weighted moving average method: , in, For the node at time The real-time failure probability, This represents the smoothed failure probability. For smoothing coefficients; Calculate the residual between the real-time failure probability and the smoothing result: , When the absolute value of the residual exceeds the residual threshold determined based on historical normal operation data, it is determined that the node has experienced a sudden increase or abnormal jump in the probability of failure, which is used to identify sudden risk changes that occur in a short period of time. The residual sequence is accumulated using a cumulative sum and trend detection method: , in, For a moment The cumulative trend statistic For the allowable normal fluctuation parameters, when When the preset trend threshold is exceeded, it is determined that the failure probability of the node has a continuous upward trend; An adaptive risk threshold is set based on the failure probability distribution of each node during its historical normal operation phases. For the i-th node, its risk threshold is... Represented as: in, This represents the first node in the historical failure probability sequence. quantiles; The current failure probability of the node, the cumulative and trend detection results, and the residual abnormal results are comprehensively judged.

[0015] The present invention, by adopting the above technical solution, has the following advantages: This invention effectively avoids the shortcomings of traditional single-layer models, such as dimensional explosion, complex modeling, and insufficient accuracy. It can adapt to the complex working conditions of frequent model changes and dynamic process adjustments in mixed-line production, significantly improving the model's generalization ability and online real-time inference efficiency. It can comprehensively cover explicit faults and early-stage hidden risks without increasing manual analysis costs, accurately locating the root cause of faults. Compared to existing technologies, this invention achieves a technological upgrade from "post-event alarm for single-point faults" to "pre-event prediction of layered risks and quantitative control throughout the entire process," effectively reducing the false detection and missed detection rates of faults in mixed-line manufacturing, reducing unplanned downtime losses and batch production quality defects, and significantly improving the operational reliability, fault management accuracy, and intelligent operation and maintenance automation level of mixed-line manufacturing. Attached Figure Description

[0016] Figure 1 This is a general flowchart of a reliability testing method for real-time mixed-line production lines according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a two-layer Bayesian network framework for real-time mixed-line production line reliability testing. Figure 3 This is a schematic diagram of the failure probability propagation of the AGV subsystem; Figure 4 This is a schematic diagram of the failure probability propagation of the CNC subsystem; Figure 5 This is a schematic diagram of the failure probability propagation of the assembly subsystem; Figure 6 This is a schematic diagram illustrating the failure probability propagation of a gantry robot subsystem; Figure 7 This is a schematic diagram illustrating the failure probability propagation of the loading and unloading subsystem; Figure 8 This is a schematic diagram of the failure probability propagation of the MES subsystem; Figure 9 This is a processing status diagram of a mixed-line manufacturing production line; Figure 10 It is a curve showing the overall failure probability of a mixed-line production line under different processing conditions. Figure 11 It is a residual peak density curve of the docking failure probability detected in real time. Detailed Implementation

[0017] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings to provide a clearer understanding of the purpose, features, and advantages of the present invention. It should be understood that the embodiments shown in the drawings are not intended to limit the scope of the present invention, but are merely illustrative of the essential spirit of the technical solution of the present invention.

[0018] In the following description, certain specific details are set forth for the purpose of illustrating various disclosed embodiments in order to provide a thorough understanding of the various disclosed embodiments. However, those skilled in the art will recognize that embodiments may be practiced without one or more of these specific details. In other instances, well-known apparatuses, structures, and techniques associated with this application may not have been shown or described in detail to avoid unnecessarily obscuring the description of the embodiments.

[0019] Unless the context requires otherwise, throughout the specification and claims, the word “comprising” and its variations, such as “including” and “having”, shall be understood to have an open, inclusive meaning, that is, to be interpreted as “including, but not limited to”.

[0020] Throughout this specification, references to "an embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Therefore, the appearance of "in an embodiment" or "an embodiment" in various places throughout the specification does not necessarily refer to the same embodiment. Furthermore, a particular feature, structure, or characteristic may be combined in any manner in one or more embodiments.

[0021] The singular forms “a” and “the” used in this specification and the appended claims include plural references unless otherwise expressly stated herein. It should be noted that the term “or” is generally used to mean “and / or” unless otherwise expressly stated herein.

[0022] like Figure 1 As shown, a reliability testing method for real-time mixed-line manufacturing production lines includes the following steps: S1. Construct a two-layer Bayesian network framework: Obtain historical operating conditions and fault data of mixed-line production lines, analyze and statistically analyze key failure factors, and then construct a layered Bayesian sub-network for failure of the bottom equipment subsystem and a Bayesian main network for the propagation of the entire line process.

[0023] Based on the summary table of operational issues and rectifications collected from the actual intelligent manufacturing integration verification line for automotive powertrains, and analyzing a total of one year's worth of failure point data, containing approximately 1400 failure events / rectification records, a two-layer Bayesian network framework for reliability testing of real-time mixed-line production lines was constructed. Its specific topology is as follows: Figure 1As shown, the system is divided into a bottom-level equipment subsystem failure association Bayesian subnetwork and an upper-level whole-line process propagation Bayesian main network. By using a two-layer network to decouple the coupling interference between single-machine system failures in multiple processes, the bottom-level subnetwork accurately identifies key failure factors in each equipment subsystem, enabling precise calculation of single-machine failure probabilities. Simultaneously, the upper-level main network quantifies the fault propagation paths, coupling weights, and diffusion strength between subsystems, clearly depicting the whole-line failure propagation mechanism. Furthermore, based on the probabilistic reasoning capabilities of the two-layer Bayesian network, the failure probabilities of production line equipment and the entire line are dynamically updated in real time, predicting the evolution trend of fault risks.

[0024] S2. Based on the probabilistic reasoning capability of a two-layer Bayesian network, dynamically and in real-time update the failure probability of equipment subsystems and the entire production line.

[0025] For each subsystem, its failure rate is calculated and updated in real time by a Bayesian network. For each subsystem, the network consists of root causes of failure, failure modes, and top event nodes.

[0026] Root cause of failure: The root cause leads to the specific failure manifestation, that is, the contents recorded in the failure list and the source of the problem; Failure Mode: This generally refers to a partial failure of a subsystem caused by multiple root causes. Taking an AGV (Automated Guided Vehicle) system as an example, this could include: scheduling and communication failure, mechanical failure, sensing and positioning failure, and docking failure. The failure rate is updated directly through real-time time-series data, and subsystem failure analysis begins with these types of failure points.

[0027] Top event: This is generally a subsystem failure behavior that is directly triggered. The overall subsystem failure rate is obtained by combining the failure rates of multiple top events. Taking an AGV (Automated Guided Vehicle) system as an example, it can include task interruption and inability to start.

[0028] Taking the automated guided vehicle (AGV) system as an example, the first step is to statistically analyze historical failure events based on the production line failure point table. To reduce the impact of occasional records and abnormal data on the modeling results, the original failure records can be filtered based on the frequency, duration, and repetition of failures, retaining failure events with a certain degree of repetition and statistical significance. Let... Indicates the deadline The cumulative number of valid failures recorded. Then, starting from time... Starting from, the length is Equivalent failure probability within the statistical window : Unified equivalent failure rate conversion window ; : Cumulative number of failures / cumulative failure rate function.

[0029] Mechanical failures generally follow a Weibull distribution, which... Instantaneous failure probability at time 1 The calculation formula is: Scale parameters ; Shape parameters This indicates that the failure rate increases over time.

[0030] For scheduling communication failures, a Goel–Okumoto distribution is generally followed, which occurs at time [time value missing]. After that, the length is average failure rate formula within the time window for: Ultimately, the total number of faults can be exposed, for example, ; : Fault exposure rate (the higher the rate, the faster the decay), for example, .

[0031] For sensing and localization failures, a Cox-Weibull distribution is generally followed, which, given the operating condition variables... Under these conditions, the sensing and positioning system at any time Instantaneous failure probability The calculation formula is: The baseline parameters are: Baseline scale parameters , Baseline shape parameters covariates are The variable takes the following values: speed ; load ; Dust levels ; Reflectivity ; Positioning residual .

[0032] Covariate coefficients for: Speed ​​coefficient ; Load factor ; Dust level coefficient ; Reflectance coefficient ; Location residual coefficient .

[0033] For docking failures, the general distribution follows the Crow-AMSAA distribution, which under a given operating condition... ,time The length after is Average docking failure probability within the time window for: in, covariates The variable takes the following values: Positioning error; Posture deviation; docking frequency; Fixture consistency deviation; Landmark recognition bias; Docking tolerance variation.

[0034] The basic parameters of the distribution are: Scale parameters; Trend parameters; Covariate coefficient vector.

[0035] Covariate coefficients for: Positioning error coefficient; Attitude deviation coefficient; Docking frequency coefficient; Fixture consistency deviation coefficient; Landmark identification deviation coefficient; Docking tolerance variation coefficient.

[0036] For failure modes in different subsystems other than AGVs, if the failure rate model name is the same, then the failure rate distribution formula type is the same, although the specific parameters may differ slightly. Failure rate models mainly include waveform failure rate models and three-stage stepped failure probability models, etc.

[0037] Among them, the waveform failure rate model is used to describe failure modes with periodic variation characteristics, including two types: sine waveform and sawtooth waveform.

[0038] A sine wave is used to describe task scheduling failure, and its formula is: in, Represents normalized time, from the current time. With total time length The ratio is calculated; This represents the linear trend term at time t. and represents the initial and final values ​​of the basic failure rate, respectively; f represents the number of cycles of the waveform over the entire time range, and within the interval Values ​​are distributed; A represents the waveform amplitude, and is within the interval... Values ​​are distributed; Indicates the initial phase of the waveform, and within the interval The values ​​are taken according to a uniform distribution. This represents the failure probability at time t.

[0039] The sawtooth waveform is used to describe data consistency failure (DCF) and data acquisition failure (DAF), and its formula is: in, Indicates time The linear trend term, and These represent the initial and final values ​​of the basic failure rate, respectively; This represents a piecewise linear sawtooth offset sequence, constructed as follows: [The time interval is...] Divided into Each segment This indicates the number of inflection points obtained by sampling within a preset range according to a uniform distribution. Indicates the first The initial length of each segment, and satisfying After generating the length of each segment, normalization is performed. Indicates the first The offset of each turning point, and according to Calculation, where This represents the fundamental amplitude of the sawtooth waveform. This represents a random adjustment coefficient indicating the magnitude of the node offset. Offset from adjacent nodes Obtained by linear interpolation; Indicates time The probability of failure.

[0040] Different equipment subsystems and failure modes differ in terms of basic failure rate level, trend and fluctuation characteristics. The parameter ranges for the starting value of basic failure rate, the ending value of basic failure rate and the fluctuation amplitude are set respectively.

[0041] Assembly system parameters:

[0042] CNC system parameters:

[0043] Parameters of the gantry robot system:

[0044] Parameters of the loading and unloading conveyor system:

[0045] For mechanical failures, sensing and positioning failures, communication failures, docking failures, and quality management failures, a three-stage stepped failure probability model can be used: in, This indicates the first breakpoint, which lies within the interval [0, 1]. The values ​​are taken according to a uniform distribution. This indicates the second breakpoint, which lies within the interval [0, 1]. The values ​​are taken according to a uniform distribution. This represents the failure probability in the first time period, which is within the interval. The values ​​are taken according to a uniform distribution. This represents the failure probability in the second time period, which is determined by... Add interval The value is obtained by uniformly randomizing the values ​​within the range; This represents the failure probability in the third time period, which is determined by... Add interval The values ​​are obtained by uniformly and randomly selecting values ​​within the range.

[0046] For failure propagation probability, the conditional probability table (CPT) of intermediate nodes can be determined using the failure history data of the production line subsystem based on the maximum likelihood estimation (MLE).

[0047] To estimate CPT using maximum likelihood estimation (MLE), follow these steps: 1) Construct the initial structure of CPT: determine the parent node of each node and determine the dimensions of the CPT table.

[0048] 2) Calculate the likelihood function: For each CPT, calculate the probability of the observed data occurring under given parameters. The likelihood function is the product of the joint probabilities of all observed data points.

[0049] 3) Maximizing the likelihood function: Adjust the parameters in CPT to maximize the likelihood function. This invention optimizes the likelihood function by taking the logarithm. The formula is shown below: in: Given parameters At that time, the likelihood function of all historical observation data; : The log-likelihood function corresponding to the likelihood function; The total number of historical observation samples; : No. Index of each observation sample; : No. The combination of parent node states of the target node in each sample; : No. The observation status of the target node in each sample; : When the parent node is in a certain state Under the condition that the target node is in state The conditional probability; The set of parameters to be estimated in the conditional probability table of a Bayesian network; : The optimal conditional probability parameters obtained through maximum likelihood estimation; The parameter value that makes the log-likelihood function reach its maximum value.

[0050] Calculation of overall production line failure probability. For a specific mixed-line manufacturing line, assume that there exists... N Each is a single-machine subsystem, and the processing status of this mixed-line production line includes... K If a certain type of product is used, then a specific product on that production line... t The state corresponding to time is . Figure 9 A processing state diagram of a mixed-line manufacturing production line is shown. Assume that for the i-th subsystem, its corresponding state participation variables are: The overall failure probability of the entire mixed-line manufacturing production line can be expressed as: This indicates the subsystem's working state. Participation status; The failure probability of the current subsystem is obtained through the CPT mentioned above; It is important to note that the failure assumptions of the subsystems are independent of each other.

[0051] S3. Reliability Assessment: The reliability of mixed-line production lines is assessed based on the failure probability of equipment subsystems and the entire line.

[0052] Based on the real-time failure probabilities of each failure root cause node, failure mode node, equipment subsystem node, and line node obtained from the above two-layer Bayesian network calculation, risk assessment and graded early warning are further performed on the mixed-line production line.

[0053] First, to reduce the impact of sensor noise and instantaneous data fluctuations on risk assessment, an exponentially weighted moving average method is used to smooth the real-time failure probability sequence of each node: in, For the node at time The real-time failure probability, This represents the smoothed failure probability. This is the smoothing coefficient. Further calculations are performed to determine the residual between the real-time failure probability and the smoothing result: When the absolute value of the residual exceeds the residual threshold determined based on historical normal operation data, it is determined that the node has experienced a sudden increase or abnormal jump in the probability of failure, which is used to identify sudden risk changes that occur in a short period of time.

[0054] Secondly, to identify situations where the failure probability continues to rise but has not yet experienced a significant jump, a cumulative sum and trend detection method is used to perform cumulative calculations on the residual sequence: in, For a moment The cumulative trend statistic This is the parameter for acceptable normal fluctuations. When When the preset trend threshold is exceeded, the failure probability of the node is determined to have a continuously increasing trend. For the root cause and failure mode nodes at the bottom level, more stringent trend detection parameters can be used to identify sharp increases and jumps in the probability of local failures; for the mid-level equipment subsystem nodes and the upper-level line nodes, relatively lenient trend detection parameters can be used to identify the risk response formed by the combined effect of multiple bottom-level nodes.

[0055] Simultaneously, an adaptive risk threshold is set based on the failure probability distribution of each node during its historical normal operation phases. For the i-th node, its risk threshold can be expressed as: in, This represents the first node in the historical failure probability sequence. Quantiles. The 95th quantile can be used as the risk threshold for bottom-level nodes, while the 90th quantile can be used for middle- and upper-level nodes, in order to balance the accuracy of bottom-level anomaly identification with the timeliness of upper-level risk response.

[0056] Finally, the current failure probability, cumulative trend detection results, and residual anomaly results of each node are comprehensively assessed. When the failure probability of a node exceeds a risk threshold, the node is determined to be in a high-risk state; when the cumulative trend statistic exceeds a trend threshold, the node is determined to be in a continuously rising risk state; when the residual exceeds a residual threshold, the node is determined to have experienced a sudden anomaly. If multiple conditions are met simultaneously, the warning level of the node is increased. Corresponding warning results are generated based on the risk level, failure root cause ranking, and fault propagation path. The warning results include at least high-risk equipment or processes, main failure modes, possible failure triggers, potentially affected equipment, and risk levels, and serve as a quantitative basis for equipment inspection sequence, preventive maintenance priority, and production scheduling intervention. When the risk level is high, priority can be given to inspecting root cause nodes with high posterior failure probabilities, and subsequent equipment and processes that may be affected by failure propagation can be preemptively investigated or scheduled for adjustment, thereby realizing the transformation from real-time failure probability calculation to production line risk assessment and fault warning.

[0057] This invention employs a hierarchical modeling framework of "lower-level equipment system sub-network failure analysis + upper-level production line main network propagation quantification + two-layer collaborative probabilistic reasoning" to decompose the overall failure problem of complex mixed-line manufacturing production lines into two independent logics: micro-failure evolution within equipment and macro-failure propagation between processes. The lower-level Bayesian sub-network isolates multi-process coupling interference, accurately identifying key failure factors in each subsystem and quantifying the failure probability of individual equipment. Furthermore, the upper-level Bayesian main network constructs a cross-system, cross-process fault propagation topology, quantifying fault propagation paths, propagation weights, and coupling influence strength, achieving visualization and refined characterization of the failure propagation mechanism. Moreover, relying on the dynamic probabilistic reasoning capabilities of the two-layer Bayesian network, this invention can dynamically update the failure probabilities at each level based on real-time production line operation data, continuously extrapolating the spread of single-point failures and the overall line risk evolution trend, achieving early fault warning, precise root cause tracing, and dynamic assessment of overall line reliability. This layered decoupling and collaborative reasoning capability effectively avoids the shortcomings of traditional single-layer models, such as dimensional explosion, mixed modeling, and insufficient accuracy. It can adapt to the complex working conditions of frequent model changes and dynamic process adjustments in mixed-line production, significantly improving the model's generalization ability and online real-time reasoning efficiency. It can comprehensively cover explicit faults and early hidden risks without increasing manual analysis costs, and accurately locate the root cause of faults. Compared with existing technologies, this invention achieves a technological upgrade from "post-event alarm for single-point faults" to "pre-event prediction of layered risks and quantitative control throughout the entire process," effectively reducing the false detection and missed detection rates of faults in mixed-line manufacturing, reducing unplanned downtime losses and batch production quality defects, and significantly improving the operational reliability, fault management accuracy, and intelligent operation and maintenance automation level of mixed-line manufacturing.

[0058] Example Based on the solution proposed in this invention, combined with the summary table of rectification of mixed-line production line operation problems, the failure point table data for a total of one year was analyzed, and the corresponding CPT table for each subsystem was calculated to obtain the failure network and failure probability of each subsystem, such as... Figures 3-8 As shown. To save space, only the corresponding CPT table for the AGV subsystem is shown below: Table 1. CPT Table for AGV Subsystem Task Interruption

[0059] Table 2. CPT Table for AGV Subsystem Failure to Start

[0060] Table 3 CPT Table for AGV Subsystem

[0061] For this actual production line, the corresponding processing status is as follows: Figure 9As shown, the real-time failure rate of the production line changes with the switching of its actual processing state, as follows: Figure 10 As shown.

[0062] Based on the above failure probability calculation results, further risk identification was performed using exponentially weighted moving average, quantile threshold, and residual peak density. The results show that the docking failure probability fluctuates significantly after approximately 3000 time steps, and exceeds the risk threshold multiple times between 3500 and 4100 time steps. Figure 11 As shown; meanwhile, the residual peak density within this interval consistently exceeds the set threshold, indicating that the anomaly is not an isolated fluctuation but rather a continuous risk. Therefore, this time interval is designated as a key warning interval, and priority is given to checking factors such as positioning error, attitude deviation, and fixture consistency.

[0063] The preferred embodiments of the present invention have been described in detail above. However, it should be understood that after reading the above teachings, those skilled in the art can make various alterations or modifications to the present invention. These equivalent forms also fall within the scope defined by the appended claims.

Claims

1. A reliability testing method for real-time mixed-line manufacturing production lines, characterized in that, Includes the following steps: S1. Obtain historical operating conditions and fault data of mixed-line production lines, analyze and statistically analyze key failure factors, and then construct a layered Bayesian sub-network of failures in the underlying equipment subsystems and a Bayesian main network of process propagation in the upper-level whole line. S2. Based on the probabilistic reasoning capability of a two-layer Bayesian network, dynamically and in real-time update the failure probability of equipment subsystems and the entire production line; S3. Evaluate the reliability of mixed-line production lines based on the failure probabilities of equipment subsystems and the entire line.

2. The reliability testing method for real-time mixed-line production lines according to claim 1, characterized in that, In S1, the historical operating conditions and fault data are a table of failure points for one year.

3. The reliability testing method for real-time mixed-line production lines according to claim 2, characterized in that, The underlying device subsystem failure Bayesian subnetwork consists of failure root causes, failure modes, and top event nodes. The failure root causes are the contents recorded in the failure list and the source of the problem. The failure modes are the cases where the subsystem partially fails due to multiple failure root causes. The top event nodes are the subsystem failure behaviors that are directly triggered.

4. The reliability testing method for real-time mixed-line production lines according to claim 3, characterized in that, The failure modes include mechanical failure, scheduling and communication failure, sensing and positioning failure, docking failure, task scheduling failure, communication failure, data acquisition failure, data consistency failure, and quality management failure.

5. The reliability testing method for real-time mixed-line production lines according to claim 4, characterized in that, The instantaneous failure rate of the mechanical failure for: ， in, For scale parameters; For shape parameters; The average failure rate of the scheduling communication failure for: ， in, To determine the total number of faults that can ultimately be exposed, For the fault exposure rate, The duration; Instantaneous failure rate of sensing and positioning systems for: ， ， in, For baseline scale parameters, The baseline shape parameter is , and the covariates are . , speed, For load, Dust level, Reflective intensity, To locate the residuals, the covariate coefficients are: ; Average failure probability of docking failure for: ， Among them, covariates ,in, For positioning error, For attitude deviation, For docking frequency, For fixture consistency deviation, Due to landmark identification bias, This refers to the tolerance variation during docking; for Scale parameters; For trend parameters; The duration; This is the vector of covariate coefficients.

6. The reliability testing method for real-time mixed-line production lines according to claim 5, characterized in that, The failure probability of the top event node is calculated using the conditional probability table CPT based on the maximum likelihood estimation (MLE).

7. The reliability testing method for real-time mixed-line production lines according to claim 6, characterized in that, The calculation of the conditional probability table (CPT) based on the maximum likelihood estimation (MLE) specifically includes: Construct the initial structure of CPT: determine the parent node of each node and the dimensions of the CPT table; Calculate the likelihood function: For each CPT, calculate the probability of the observed data occurring given the parameters. The likelihood function is the product of the joint probabilities of all observed data points. ;in, For given parameters At that time, the likelihood function of all historical observation data; This represents the total number of historical observation samples. For the first Index of each observation sample; For the first The combination of parent node states of the target node in each sample; For the first The observation status of the target node in each sample; For the parent node to be in state Under the condition that the target node is in state The conditional probability; This is the set of parameters to be estimated in the conditional probability table of the Bayesian network; To maximize the likelihood function: Adjust the parameters in CPT to maximize the likelihood function. ,in, Let log-likelihood function be the function corresponding to the likelihood function. ; These are the optimal conditional probability parameters obtained through maximum likelihood estimation.

8. The reliability testing method for real-time mixed-line production lines according to claim 6, characterized in that, The failure probability of the device subsystem is calculated using the same method as the failure probability of the top event node.

9. The reliability testing method for real-time mixed-line production lines according to claim 1, characterized in that, In S2, the probability of line failure is expressed as: ， in, This indicates the working state of the i-th subsystem. The participation status is indicated by 0 for no participation and 1 for participation; This represents the failure probability of the i-th subsystem.

10. The reliability testing method for real-time mixed-line production lines according to claim 1, characterized in that, S3 include: The real-time failure probability sequence of each node is smoothed using an exponentially weighted moving average method: ， in, For the node at time The real-time failure probability, This represents the smoothed failure probability. For smoothing coefficients; Calculate the residual between the real-time failure probability and the smoothing result: ， When the absolute value of the residual exceeds the residual threshold determined based on historical normal operation data, it is determined that the node has experienced a sudden increase or abnormal jump in the probability of failure, which is used to identify sudden risk changes that occur in a short period of time. The residual sequence is accumulated using a cumulative sum and trend detection method: ， in, For a moment The cumulative trend statistic For the allowable normal fluctuation parameters, when When the preset trend threshold is exceeded, it is determined that the failure probability of the node has a continuous upward trend; An adaptive risk threshold is set based on the failure probability distribution of each node during its historical normal operation phases. For the i-th node, its risk threshold is... Represented as: in, This represents the first node in the historical failure probability sequence. quantiles; The current failure probability of the node, the cumulative and trend detection results, and the residual abnormal results are comprehensively judged.

Images