Automobile module production and detection integrated assembly line control method

Abstract

This invention discloses an integrated production and testing line control method for automotive modules, belonging to the field of industrial automation testing technology. It encapsulates vehicle transit information as event objects, obtains the vehicle's unique identification code, workstation identifier, acquisition terminal type, and on-site and server timestamps, and calculates the observation delay. It recursively generates a delay confidence bound by filtering the delay at each workshop and writes it into a parameter pool. Based on the delay confidence bound, it calculates the over-boundary quantity and delay attenuation factor, constructs a hidden Markov model of the vehicle's logical state for state probability inference, and performs evidence fusion to output the vehicle's logical state, confidence level, and conflict level when multiple terminals conflict. It determines scheduling feasibility based on robust statistical thresholds of confidence level and conflict level, and the routing controller issues transmission control actions to the programmable logic controller. It uses physical arrival feedback to form an arrival deviation sequence and performs accumulation and change point detection, updating the delay model and state model parameters upon triggering.

Description

Technical Field

[0001] This invention relates to the field of industrial automation testing technology, and more specifically, to a method for integrated production line control of automotive module production and testing. Background Technology

[0002] In automotive module production lines, vehicle transit information is typically acquired through fixed identification terminals, RFID reading and writing terminals, handheld identification terminals, or industrial tablets. The control system aggregates and stores the acquired events using the vehicle's unique identification code and workstation identifier as indexes. The routing controller then calculates the target workstation or buffer area by combining process and topology parameter libraries. Finally, the programmable logic controller executes control actions such as turnout switching, release, and conveying.

[0003] The existing technology has the following shortcomings:

[0004] Multi-source data acquisition is prone to event loss, out-of-order processing, and conflicts under different workshop and link conditions, leading to jumps in vehicle position or stage information. When the routing controller directly uses such inputs for path calculation, it may cause vehicles to be sent to process sections or buffer areas that should not be entered. At the same time, changes in the environment and equipment status during long-term operation will cause drift in latency levels and observation characteristics. If there is no online drift identification and parameter update mechanism based on physical position feedback, the model will gradually become mismatched with the field.

[0005] To address the above problems, this invention proposes a solution. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide an integrated production and testing line control method for automotive modules to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for controlling an integrated production and testing line for automotive modules, including the following steps;

[0009] Step S1: Construct event objects and calculate observation delays. Perform state estimation for the welding and painting workshops using a Kalman filter. Initialize filter parameters using the observation delay sequence, median, and median absolute deviation. Combine the dynamic estimation of process noise and observation noise to recursively calculate the equivalent delay mean and estimation uncertainty, thereby generating the delay confidence bound for each workshop.

[0010] Step S2: Calculate the superboundary quantity of the event based on the time delay confidence bound, construct the time delay attenuation factor, construct the hidden Markov model to perform probabilistic inference of the vehicle logic state, integrate the attenuation factor into the transmission probability to achieve dynamic adjustment, and introduce evidence theory fusion in the multi-terminal conflict scenario. Construct the evidence reliability based on the terminal reliability coefficient and the attenuation factor, and output the vehicle logic state, confidence and conflict degree.

[0011] Step S3: Using the confidence and conflict history window, construct an adaptive threshold by the median and the absolute deviation of the median to determine the feasibility of the vehicle status. If the threshold is met, input the route controller to perform path calculation and issue instructions; otherwise, perform abnormal traffic diversion and push alarm information.

[0012] Step S4: Construct an arrival deviation sequence based on the physical arrival timestamp and the predicted arrival timestamp fed back by the programmable logic controller, detect structural drift by the cumulative sum method, amplify and reset the process noise variance and posterior error covariance of the Kalman filter after triggering, retrain the parameters of the hidden Markov model, and dynamically adjust the reliability coefficient of the acquisition terminal.

[0013] In a preferred embodiment, step S1 includes the following:

[0014] The median and median absolute deviation are calculated based on historical observation time delay sequences. The median is used as the initial state estimate of the Kalman filter, and the median absolute deviation is used as the initial covariance estimate of the Kalman filter.

[0015] During the Kalman filter recursion process, the statistical characteristics of the process noise and observation noise are updated using the current observation delay, and the equivalent delay mean and estimation uncertainty are corrected in real time based on the updated statistical characteristics.

[0016] In a preferred embodiment, step S2 includes the following:

[0017] Calculate the degree to which the event observation delay exceeds its delay confidence bound, and obtain the delay attenuation factor based on the degree of exceedance. The attenuation factor is used to characterize the confidence of the observation value.

[0018] In the calculation of the emission probability of the Hidden Markov Model, a time delay attenuation factor is introduced so that the contribution of the current observation to the state estimation decreases as the attenuation factor decreases, thereby realizing the dynamic adjustment of the emission probability.

[0019] For conflicting information reported simultaneously by multiple acquisition terminals, the basic probability allocation of evidence from each terminal is determined based on the reliability coefficient and real-time attenuation factor of each acquisition terminal. Evidence theory is used to fuse multiple pieces of evidence, and the fused vehicle logic state, confidence level and conflict level are output.

[0020] In a preferred embodiment, step S3 includes the following:

[0021] The confidence and conflict levels of the most recent preset number of moments are collected to form a historical window. The median and absolute deviation of the median confidence level within the window are calculated to generate a confidence threshold. The median and absolute deviation of the median conflict level within the window are calculated to generate a conflict threshold.

[0022] The confidence level at the current moment is compared with the confidence level threshold, and the conflict level at the current moment is compared with the conflict level threshold. If the confidence level is higher than the confidence level threshold and the conflict level is lower than the conflict level threshold, it is determined to be feasible. The vehicle status is input into the routing controller to perform path calculation and issue instructions. Otherwise, it is determined to be infeasible. Abnormal traffic diversion is performed and alarm information is pushed.

[0023] In a preferred embodiment, step S4 includes the following:

[0024] For each workstation, the difference between the physical arrival time stamp fed back by the programmable logic controller and the pre-predicted arrival time stamp is recorded to form an arrival deviation time series;

[0025] The cumulative sum method is used to monitor whether the mean of the position deviation time series has undergone structural drift. When drift is detected, the process noise variance of the Kalman filter is increased and the posterior error covariance is reset. The parameters of the hidden Markov model are retrained using the data of the current time period, and the reliability coefficient of the acquisition terminal is adjusted according to the degree of drift.

[0026] The technical effects and advantages of the integrated production and testing assembly line control method for automotive modules of this invention are as follows:

[0027] By independently maintaining and observing delays at each workshop and outputting delay confidence bounds, gating and attenuating the contribution of late and out-of-order data collection events are achieved, reducing the impact of low-quality inputs on subsequent inferences. Through probabilistic inference of vehicle logic states and fusion of evidence from multiple terminals, confidence and conflict levels are output simultaneously, enabling the vehicle state description to have measurable reliability and consistency. A feasibility judgment checkpoint is introduced before the routing controller performs path calculations, and robust statistics of confidence and conflict levels are used to adaptively form thresholds, reducing the probability of abnormal states directly driving physical scheduling. The physical arrival feedback of the programmable logic controller is used to construct arrival deviations and perform change point detection. When drift is triggered, the parameters of the delay model and state model are updated in a closed loop, which facilitates maintaining the consistency between the model and the field under long-term operating conditions. Attached Figure Description

[0028] Figure 1 A schematic diagram illustrating the structure of an integrated production and testing assembly line control method for automotive modules.

[0029] Figure 2This is a schematic diagram of the integrated production and testing assembly line control method for automotive modules according to the present invention.

[0030] Figure 3 A flowchart is generated for the trusted bounds of the two-workshop delay. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Example

[0033] Please see Figures 1-3 As shown, this invention discloses an integrated production and testing assembly line control method for automotive modules, including the following steps:

[0034] Step S1: Construct event objects and calculate observation delays. Perform state estimation for the welding and painting workshops using a Kalman filter. Initialize filter parameters using the observation delay sequence, median, and median absolute deviation. Combine the dynamic estimation of process noise and observation noise to recursively calculate the equivalent delay mean and estimation uncertainty, thereby generating the delay confidence bound for each workshop.

[0035] Step S2: Calculate the superboundary quantity of the event based on the time delay confidence bound, construct the time delay attenuation factor, construct the hidden Markov model to perform probabilistic inference of the vehicle logic state, integrate the attenuation factor into the transmission probability to achieve dynamic adjustment, and introduce evidence theory fusion in the multi-terminal conflict scenario. Construct the evidence reliability based on the terminal reliability coefficient and the attenuation factor, and output the vehicle logic state, confidence and conflict degree.

[0036] Step S3: Using the confidence and conflict history window, construct an adaptive threshold by the median and the absolute deviation of the median to determine the feasibility of the vehicle status. If the threshold is met, input the route controller to perform path calculation and issue instructions; otherwise, perform abnormal traffic diversion and push alarm information.

[0037] Step S4: Construct an arrival deviation sequence based on the physical arrival timestamp and the predicted arrival timestamp fed back by the programmable logic controller, detect structural drift by the cumulative sum method, amplify and reset the process noise variance and posterior error covariance of the Kalman filter after triggering, retrain the parameters of the hidden Markov model, and dynamically adjust the reliability coefficient of the acquisition terminal.

[0038] In step S1, event objects are constructed and observation delays are calculated. State estimations are performed on the welding and painting workshops using a Kalman filter. Filter parameters are initialized using the observation delay sequence, median, and median absolute deviation. Combined with dynamic estimations of process noise and observation noise, the equivalent delay mean and estimation uncertainty are recursively calculated, thereby generating the delay confidence bounds for each workshop. Specific details include:

[0039] Each time a vehicle is identified and transmitted by the data collection terminal, the transit event is defined as an event object, denoted as . The event object includes the following fields: vehicle unique identifier, workstation identifier, data acquisition terminal type identifier, on-site event timestamp, and server reception timestamp, denoted as... ;in, For the first The transit event object, A unique identification code for the vehicle. For workstation identification, For identifying the type of data acquisition terminal, This refers to the timestamp of the on-site event recorded by the data acquisition terminal when it completes the identification process on-site. The server-side data receiver completes the parsing of the server-side data receiving timestamp recorded when it is enqueued. The on-site event timestamp and the server receiving timestamp are kept on the same time base through a unified time synchronization service. The time synchronization method can be a network time synchronization service or an equivalent unified time synchronization service. The original value of the on-site event timestamp is retained for delay calculation.

[0040] The workstations are pre-configured into welding workshop workstation sets according to their workshop affiliation. Grouping with painting workshop workstations ,when At that time, the incident belonged to the welding workshop; when At that time, the incident belonged to the painting workshop, and the workshop identification was defined. ;

[0041] For any event object, calculate its observation delay, which is determined by the timestamp received by the server for the k-th event. The timestamp of the event at the scene of the kth event We obtain the difference, and denote the observation delay of the event arriving at the kth time as . ;

[0042] Each workshop maintains its own observation delay cache queue, statistical sliding window, and model parameter pool. The observation delay cache queue is used to store the most recent... The observation delay is used to store the most recent L logical state quality indicators, and the model parameter pool is used to publish the model parameters that need to be called in subsequent steps.

[0043] For each event object First determine its workstation identification. Does it belong to the welding workshop workstation set? Or a collection of workstations in the painting workshop Obtain workshop identification The observation delay is obtained by subtracting the server's received timestamp from the on-site event timestamp. and will , Write the observation delay buffer queue of the workshop;

[0044] The observation time delay sequence of the welding workshop and the painting workshop are respectively input into the one-dimensional Kalman filter of the workshop, and the output is the equivalent time delay mean estimate and estimation uncertainty of the workshop, and the time delay confidence bound is further generated.

[0045] Taking the welding workshop as an example, define state variables. This is the average equivalent perception delay of the welding workshop at the k-th update. The workshop network and terminal processing delays drift slowly over a short period of time, and a first-order random walk is used as the state evolution hypothesis. ;in, Let be the average equivalent sensing delay of the welding workshop at the k-th update. Let be the average equivalent sensing delay of the welding workshop at the (k-1)th update. This is the process noise term in the welding workshop, used to describe the random drift of the true mean time delay.

[0046] The observation equation for the welding workshop is defined as: the average equivalent sensing delay of the welding workshop at the k-th update. Noise level observed in the welding workshop The sum is denoted as the observation delay input of the welding workshop at the k-th update. ;

[0047] The painting workshop adopts a isomorphic definition, and is respectively based on , , as well as This indicates the corresponding quantity. The filters in the two workshops are independent of each other to avoid the statistical characteristics of one workshop affecting the other.

[0048] Collect continuous for each workshop The observation delays are used as a cold start sample set; examples can be taken. The cold start sample set of the welding workshop is recorded as follows: The cold start sample set of the painting workshop is as follows ;

[0049] The median of the cold start sample set is used as the initial state estimate of the welding workshop filter, denoted as: The same definition applies to the painting workshop. ;

[0050] For example, the median delay of 30 time delay samples observed during the cold start phase of the welding workshop is 0.180 seconds. The median time in the painting workshop is 0.260 seconds. Second;

[0051] The initial fluctuation scale is estimated using the median absolute deviation, and its square is used as the initial error covariance. First, the median absolute deviation is calculated: ;in, z is the median absolute deviation of the cold start sample set in the welding workshop; z is the set Any observation delay sample in the middle; It is an absolute value operator; For median statistic calculation;

[0052] The robust scale is defined by converting the absolute deviation of the median to a robust scale with the same dimensions as the standard deviation. : ;in, This is a robust scalar estimate for the cold start phase of the welding workshop; 1.4826 is the coefficient for converting the absolute deviation of the median to an approximate standard deviation. The median absolute deviation of the cold start sample set in the welding workshop;

[0053] Based on this, the initial error covariance is set as follows: ;in, The initial error covariance of the filter in the welding workshop; For robust metric estimation during the cold start phase of the welding workshop, the same principle applies to the painting workshop. ;

[0054] For example, if the median absolute deviation of the cold start sample in the welding workshop is 0.020 seconds, then the robustness scale... seconds, thus Square second, a unit of variance expressed in seconds;

[0055] To adapt to changes in workshop load, a sliding window is used to make robust estimates of the difference and residuals. It is assumed that the welding workshop is the closest to the... The observation delay is Examples are acceptable Where k represents the current update sequence number of the filter in the welding workshop. This represents the local sample number in the most recent Nq observation delays. ,and This only indicates the relative position of adjacent observation delays within the sliding window and does not represent the global event sequence number or global update sequence number. The difference sequence is defined as the difference between two adjacent observation delays: [Welding workshop number] Next and first The difference in the second observation delay ;

[0056] Calculate the absolute deviation of the median for the differenced series and convert it to a robust metric. Square the variance to obtain the process noise variance at the k-th update of the welding workshop. ;

[0057] Define the first sliding window in the welding workshop Measurement residuals corresponding to each observation delay It is obtained by subtracting the mean posterior delay estimate from the current observation delay and the previous time delay;

[0058] For the recent The absolute deviation of the median of the residuals was calculated, converted to a robust scale, and then squared to obtain the observation noise variance at the k-th update of the welding workshop. ;

[0059] The painting workshop obtained it using the same method. and ;

[0060] In obtaining and Afterwards, the welding workshop filter performs each new observation... Perform prediction and correction;

[0061] The prior state estimate is taken from the posterior state estimate of the previous time step: ;in, Estimate the mean of the prior delay for the k-th update in the welding workshop; For the welding workshop The mean posterior delay estimate after the next update;

[0062] Prior error covariance of the k-th update in the first welding workshop It is the sum of the posterior error covariance and the process noise variance at the previous time step;

[0063] The Kalman gain is the ratio of the prior error covariance to the sum of the prior error covariance and the observation noise variance. ;in, The Kalman gain updated for the kth time in the welding workshop; The prior error covariance for the kth update in the welding workshop; The observation noise variance for the kth update in the welding workshop;

[0064] The posterior state estimate is the prior estimate plus the Kalman gain multiplied by the innovation term, which is the difference between the observation delay and the prior estimate. ;

[0065] The posterior error covariance is: ;

[0066] The painting workshop was calculated based on the same structure. and ;

[0067] The reliability bound for delay is defined as the posterior delay mean estimate plus the uncertainty term. Therefore, the reliability bound for delay in the welding workshop is: ;in, The reliable bound of the delay output for the k-th update in the welding workshop; This is an estimate of the posterior delay mean after the k-th update in the welding workshop; where, The confidence bound coefficient is used to determine the width of the confidence range; Let be the posterior error covariance after the k-th update in the welding workshop;

[0068] The reliability boundary of the time delay in the painting workshop is: ];in, The reliable bound of the delay output for the k-th update in the painting workshop; This is an estimate of the mean posterior time delay after the kth update in the painting workshop; The confidence bound coefficient; Let be the posterior error covariance after the kth update in the painting workshop;

[0069] These are custom coefficients, which can be specified by the project configuration, and serve as a trade-off between relaxing tolerance for late data and maintaining strict gating. An example is provided. This corresponds to approximately 95% coverage under the normality assumption; it can also be adjusted within the range of 1.0 to 3.0 according to the workshop's tolerance for false positives and false negatives, and this coefficient remains unchanged once set in the same operating version;

[0070] For example, at a certain moment, the welding workshop is updated. Second, Square seconds, take ,but Seconds, therefore The value of seconds indicates that when the observation delay of an event in the welding workshop does not exceed approximately 0.2292 seconds, it can be considered to be within the current reliable delay range.

[0071] Will and The generated timestamp and version number are written into the latency parameter pool;

[0072] By using a dual-workshop Kalman filter recursive calculation, the random fluctuations of the original observation delay are transformed into a reliable upper bound parameter that can be updated over time. A reliable delay bound is output for the welding workshop and the painting workshop respectively. The next step will use this reliable delay bound as the gating basis to classify and attenuate the late data. Based on this, a probabilistic reconstruction model of the vehicle's logical state will be constructed, so that the routing controller can obtain a more stable and interpretable vehicle state input.

[0073] In step S2, based on the time delay confidence bound, the event is calculated to exceed the bound, a time delay attenuation factor is constructed, a hidden Markov model is built to probabilistically infer the vehicle's logic state, the attenuation factor is integrated into the transmission probability to achieve dynamic adjustment, and evidence theory fusion is introduced in the multi-terminal conflict scenario. Based on the terminal reliability coefficient and the attenuation factor, the evidence reliability is constructed, and the vehicle logic state, confidence level, and conflict level are output. Specific content includes:

[0074] After outputting the reliability bounds of the delay in the welding and painting workshops and quantifying the degree of lateness of each event, the system further addresses the problem of vehicle position jumps caused by the loss, out-of-order, and conflict of multi-source acquisition events. Using the observation delay of each event and the reliability bound of the corresponding workshop delay as gating inputs, the system assigns a calculable attenuation factor to the events and constructs a hidden Markov model inference process for the vehicle's logical state. Simultaneously, evidence theory fusion is introduced in the scenario of conflict between multiple acquisition terminals. Finally, the system outputs vehicle logical state, confidence, and conflict indexes that can be directly used by the routing controller. To ensure feasibility, the system clearly defines the source of model training data, training input and output, training process, and online inference process, and provides numerical examples.

[0075] For any new event, the observation delay is obtained. And read the reliability bound of the delay of the workshop to which the event belongs from the delay parameter pool, denoted as Define the superboundary quantity as: ;in, Let be the outbound quantity of the k-th event; Let be the observation delay for the k-th event; Let be the reliable bound of the time delay of the workshop to which the k-th event belongs at that moment;

[0076] Define the time delay decay factor From the recent The scale parameters for constructing the supercritical quantity sample are given, assuming the workshop is the closest The set of superboundary quantities is Examples are acceptable First, calculate the absolute deviation of the median and convert it to a robust metric: ;in, To provide a robust standard for the workshop's out-of-bounds capacity. For the set of superboundary quantities The absolute deviation of the median; For the workshop recently A set of samples exceeding the limit;

[0077] when To avoid division by zero, the lower limit of the scale is set to... For example, a time of 0.001 seconds can be used, and the effective scale can be set to... Define the attenuation factor: ;in, Let be the delay decay factor for the k-th event; It is an exponential function; Let be the outbound quantity of the k-th event; The attenuation scale parameter is determined by the robust scale and lower scale limit of the supercritical quantity.

[0078] For example, the observation delay of an event in a painting workshop =0.420 seconds, corresponding to the reliability boundary of the delay. Seconds, then the superlimit quantity seconds, if the time is calculated seconds, then This result indicates that the contribution of the late event to subsequent probability inference is significantly reduced, but it is not directly discarded;

[0079] Construct a Hidden Markov Model for the vehicle's logical state. The input to this model is: a unique identification code for the same vehicle. The sequence of observation symbols in time order; the observation symbols are mapped from the workstation category to the acquisition terminal type. The output of this model is: the probability distribution vector of the vehicle in each logical state at the current moment, and the most likely logical state and its probability can be further obtained.

[0080] Define the set of logical states as ;in, A set of vehicle logical states; Indicates the workstation's arrival status; Indicates the processing status of the workstation; Indicates that the user has left their workstation. This indicates that the system has entered a cached state. Indicates a cache wait state;

[0081] Pre-configure station category functions in the process and topology parameter library For example, workstations can be categorized into loading points, process changeover points, buffer entry points, buffer exit points, and arrival detection points, and then the types of data acquisition terminals can be further classified. Combined with workstation category, it is mapped to the observation symbol o: Where o is the observation symbol corresponding to the event; For the observation symbol mapping function; Mapping workstation identifiers to workstation categories; For event workstation identification; For identifying the type of data acquisition terminal;

[0082] Mapping function The implementation can be enumeration mapping or rule mapping. For example, when the workstation type is a cache entry and the acquisition terminal is an RFID reader / writer terminal, the observation symbol cache entry RFID observation is output. When the workstation type is a process switching point and the acquisition terminal is a wireless handheld reader terminal, the observation symbol switching point handheld reader observation is output.

[0083] Hidden Markov models include initial state distributions State transition probability matrix Basic launch probability ,in, This represents the initial state distribution composed of the initial probabilities of each logical state. This represents the initial state distribution. Corresponding logical state The probability components, the state transition probability matrix describes the transition pattern of the logic state over time, the fundamental emission probability describes the fundamental probability of producing a certain observed symbol in a certain logic state, and the time delay decay factor. The launch probability will be dynamically scaled to achieve a controllable reduction in the contribution of late data to state inference;

[0084] Acquire the arrival signal of the physical arrival detection point, and align the workstation category and time triggered by the arrival signal with the event sequence of the acquisition terminal according to the vehicle's unique identification code to form training samples;

[0085] The training input is a set of multiple vehicle observation symbol sequences. Each sequence is ;in, The sequence of observation symbols for the nth vehicle during the training period; Let be the observation symbol for the nth vehicle at the t-th observation time; Let be the total length of the observation symbols for the nth vehicle;

[0086] Optional training inputs also include partially alignable logical state label sequences inferred from in-place detection points. If complete labels cannot be obtained, unsupervised training is used.

[0087] The training output is the state transition probability matrix of the Hidden Markov Model. With base launch probability And write it into the state model parameter pool;

[0088] When partial state transitions can be deduced from the arrival detection points and process cycle time, such as the sequence of station arrival, station processing, and station departure, the frequency of adjacent state pairs can be counted. The transition probabilities are then normalized to obtain: ;in, To start from logical state Transition to logical state The transition probability; For training data from logical states arrive The transfer count; where, From The sum of transition counts from the starting point to all possible successor states;

[0089] The base emission probability is also obtained by normalizing the number of observed symbols in a certain state. ;in, For logical state The observation symbol is The base launch probability; In the training data, in the logical state The observed symbol The count; In logical state The sum of the counts of all observed symbols;

[0090] For each vehicle, a sequence of observation symbols is maintained and forward inference is performed. To reflect the gating attenuation effect, the time delay attenuation factor is... Multiplying by the base emission probability yields the dynamic emission probability, which is defined as: ;in, Let k be in the logical state at time k. The observation symbol is The dynamic emission probability; The base emission probability obtained during training; Let k be the observation symbol at time k;

[0091] Normalize the dynamic emission probabilities of each state at the same time: ;in, This represents the normalized dynamic emission probability. This represents the unnormalized dynamic emission probability. Normalization factor;

[0092] The forward inference recursion is as follows: Initial time: ;in, At time 1, the logic state is... The forward probability; Initial state distribution Corresponding logical state The probability of; where, Let be the normalized dynamic emission probability at time 1; The observation symbol for time 1;

[0093] Recursive time, i.e. : ;in, The logical state is at time t. The forward probability; For a moment In logical state The forward probability; To start from logical state arrive The transition probability; Let be the normalized dynamic emission probability at time t; The observation symbol at time t;

[0094] The forward probability is normalized to obtain the state probability distribution vector. : ;in, The vehicle is in a logical state at time t. The normalized probability; Forward probability; As the normalization factor,

[0095] For example, the probability distribution of a vehicle at time t is: , , , , If the probability is 0.70, then the vehicle is most likely in a processing state at the workstation.

[0096] When the same vehicle is observed simultaneously by multiple acquisition terminals within a short time window, and the probability distributions inferred by each terminal are different, evidence theory fusion is introduced to define the construction method of the evidence body.

[0097] Configure a reliability coefficient for each type of data acquisition terminal. The value ranges from 0 to 1 and is used to reflect the reading stability of the terminal.

[0098] Among them, the reliability coefficient is 0.95 for fixed barcode reading terminals, 0.90 for radio frequency identification reading and writing terminals, 0.85 for wireless handheld reading terminals, and 0.80 for industrial tablet data acquisition terminals. This reliability coefficient is a custom coefficient, derived from equipment acceptance data or historical missed reading rate statistics, and is consistent within the same version.

[0099] State probability distribution generated by a certain terminal Multiplying this by the reliability of the evidence yields the basic probability allocation, where the reliability of the evidence is defined as the product of the terminal reliability coefficient and the delay attenuation factor: ;in, The reliability of evidence for the k-th event; The reliability coefficient corresponding to the type of terminal used to collect this event; For the first The identifier of the terminal type for collecting each event; Let be the delay decay factor for the k-th event;

[0100] It should be noted that, in the process of constructing this piece of evidence, one of the reports submitted by the data collection terminal... Each event corresponds to one piece of evidence, therefore the first event... Article 1 and Article 2 The events have a corresponding relationship. , In All events use the same sequence number to indicate the event originating from the first event. Evidence formed by the incident For the first The reliability of evidence formed by the events;

[0101] For each single-state set The mass distribution is as follows: ;in, For the first One logical state, For the first The evidence formed by the events corresponds to the logical state. The basic probability assignment of a single-state set; For the first The reliability of evidence formed by the events; This represents the state probability corresponding to the evidence.

[0102] Remaining mass allocated to the set of total uncertainty : ;in, For the first The evidence formed by these events is assigned to the basic probability of the fully uncertain set; For the reliability of evidence;

[0103] Regarding the two pieces of evidence and The conflict coefficient is defined as: ;in, X represents the conflict coefficient between the two pieces of evidence; X and Y are set elements in the evidence theory framework, and the set elements include those corresponding to each logical state. Single-state sets and completely uncertain sets ; This represents the intersection of the set of vehicle logical states supported by the first piece of evidence and the set of vehicle logical states supported by the second piece of evidence. This indicates that the two sets of vehicle logical states supported by the evidence do not share a common logical state. and Assign values ​​to the basic probabilities on the corresponding set;

[0104] For sets With sets If the intersection is not empty, then and The product of the products is used as the composition contribution of the corresponding intersection, and is expressed as... Normalization is performed, and the normalized composite contribution is allocated to... The corresponding set of vehicle logic states is used to obtain the basic probability allocation after combining the two pieces of evidence. The normalized synthesis contribution is expressed as: ;in, The conflict coefficient between the two pieces of evidence; and For set elements in the vehicle logic state recognition framework; and Assigning basic probabilities to the two pieces of evidence on their corresponding sets;

[0105] The single state with the highest probability in the synthesized result is output as the vehicle's logical state: The state corresponding to the largest value is taken as the vehicle's logical state. And use its corresponding value as the confidence level. Simultaneously output the conflict coefficient As a conflict index, to avoid excessive conflict and numerical instability when synthesizing multiple pieces of evidence, a conflict stopping threshold can be set. The example uses 0.95. When the conflict coefficient exceeds this threshold, the synthesis stops and the current result and conflict degree are output directly.

[0106] By gating and attenuating the event observation delay, the probabilistic reconstruction of the vehicle's logical state is realized, and a quantifiable conflict index is output in multi-terminal conflict scenarios. The next step will take the output vehicle logical state, confidence level and conflict level as the core features of the scheduling input quality, and use robust statistical thresholds to determine feasibility and divert traffic, thereby avoiding the routing controller from generating erroneous physical scheduling instructions driven by low-quality data.

[0107] In step S3, using the confidence and conflict history window, an adaptive threshold is constructed based on the median and the absolute deviation of the median to determine the feasibility of the vehicle status. If the threshold is met, the route controller is input to perform path calculation and issue instructions; otherwise, abnormal traffic diversion is performed and alarm information is pushed. Specific details include:

[0108] If the input status is statistically unreliable, it may cause vehicles to be mistakenly sent to process sections or buffer areas that should not be entered. Before the routing controller performs path calculation, a scheduling feasibility judgment checkpoint is set. The robust statistics of confidence and conflict in the historical window are used to adaptively form a threshold, and then each vehicle is scored for feasibility and diverted to ensure that automatic scheduling is only executed when the input quality meets the standard.

[0109] The input is: the vehicle's most recent output from the vehicle state pool. and the workshop markings of the vehicle's most recent incident. The output is: Feasibility indicator and diversion action indicators ;

[0110] The maintenance of two types of sliding windows are divided into confidence windows for the welding workshop and the painting workshop. Conflict Window The window length is set to L, which can be set to L=100. When a new output record is generated in step S2, the confidence score of that record is written to... Write the conflict level into The window uses a circular update method to ensure that the statistics reflect the true quality level of the most recent operating phase;

[0111] A robust threshold is constructed using the absolute deviation of the median. Taking the confidence window as an example, the median is first calculated: ;in, This represents the confidence window value for the workshop. A confidence sliding window for the workshop;

[0112] Next, calculate the absolute deviation of the median: ;in, c is the absolute deviation of the confidence window of the workshop; c is the window size. For any confidence level sample, represents the middle value within the confidence window.

[0113] Define the lower confidence threshold for: ;in, This represents the lower confidence threshold for the workshop. This represents the middle value within the confidence window. This represents the confidence threshold relaxation coefficient. The absolute deviation of the confidence level window;

[0114] in This is a custom coefficient used to control the tightness of the threshold. An example is provided. When the overall perceived quality in the workshop declines, It will decrease and The threshold may rise, causing it to adaptively shift downwards to avoid widespread false rejections; when the workshop is stable, the threshold remains at a relatively high standard.

[0115] Similarly, for the conflict window, first calculate the median: ;in, This represents the median conflict level window for the workshop. A sliding window for the conflict level of the workshop;

[0116] Next, calculate the absolute deviation of the median: ;in, K represents the absolute deviation of the conflict level window in the workshop; K is the window size. Any conflict degree sample; This represents the median of the conflict level within the window.

[0117] Define the upper limit threshold for conflict degree for: ;in, The upper limit threshold for conflict level in the workshop; This represents the median of the conflict level within the window. The threshold stringency coefficient represents the conflict level. The absolute deviation of the median of the conflict degree window;

[0118] in For custom coefficients, the following example is available. This is used to remove vehicle records with significantly higher conflict levels than the current average, thus avoiding scheduling errors caused by conflicting data from multiple sources.

[0119] When the routing controller is ready to perform route calculation for a vehicle, it reads the vehicle's most recently output logical state from the vehicle state pool. Confidence level Conflict level And read from the threshold pool of the corresponding workshop and Feasibility assessment employs logic and rules: ;in, As an indicator of scheduling feasibility; is the confidence level; where, This is the lower confidence threshold for the workshop. Conflict level; This is the upper limit threshold for the conflict level in this workshop;

[0120] when If true, the vehicle's logic state will be... As the current input route controller for the vehicle, it performs path calculations by combining the topology parameter library and the process cycle parameter library. The path calculation can be implemented using a graph search algorithm or a rule engine. The output includes the target workstation or target buffer area identifier and the corresponding conveying control action. The action is then sent to the programmable logic controller to complete control operations such as switch switching, release, or writing the conveying target, and records the command issuance timestamp. ;

[0121] when If the error is false, entry into the automatic scheduling path is not permitted, and an abnormal diversion is executed. The abnormal diversion action can be configured to either lock the vehicle in place or guide it to an abnormal waiting area. Locking the vehicle in place involves sending a hold signal to the programmable logic controller (PLC) to keep the vehicle in its current position and wait for manual confirmation. Guiding the vehicle to an abnormal waiting area involves replacing the target node with the abnormal waiting area identifier. Control commands are issued in conservative mode, and alarm information is pushed to the human-machine interface. The alarm information includes at least the vehicle's unique identification code, the current vehicle logic state, confidence level, conflict level, and threshold snapshot.

[0122] If a welding workshop's vehicle outputs , The threshold is , If the feasibility flag is true, then the feasibility flag is true if another vehicle outputs... or If the feasibility flag is false, it will enter an abnormal diversion process;

[0123] By constructing a robust threshold based on the median absolute deviation, the vehicle condition quality judgment is transformed from a fixed empirical threshold to a statistical threshold that adapts to workshop operation. A data firewall is formed at the front end of the routing controller. The next step will use the deviation sequence between the physical arrival feedback after the routing controller issues the command and the predicted arrival time to detect whether structural drift occurs in the workshop operating environment or sensing link. When drift is triggered, the model parameters of steps S1 and S2 will be updated in a closed loop.

[0124] In step S4, an arrival deviation sequence is constructed based on the physical arrival timestamp fed back by the programmable logic controller and the predicted arrival timestamp. Structural drift is detected using the cumulative sum method. After triggering, the process noise variance and posterior error covariance of the Kalman filter are amplified and reset. The hidden Markov model parameters are retrained, and the reliability coefficient of the acquisition terminal is dynamically adjusted. Specific details include:

[0125] For long-term environmental changes and equipment performance degradation, such as changes in wireless coverage, aging of acquisition terminals, and changes in message queue congestion patterns, these changes will cause the time delay model in step S1 and the state model in step S2 to gradually mismatch. To establish a closed loop across the information domain to the physical domain, the physical arrival signal returned by the programmable logic controller is used as a highly reliable feedback to construct the deviation sequence between the actual arrival and the predicted arrival. Structural drift is identified through accumulation and change point detection. Once drift is triggered, the noise parameters in step S1 and the model parameters in step S2 are automatically adjusted so that convergence can be achieved without interruption.

[0126] The input is: the timestamp of the command issuance recorded when the routing controller issues the command. Target node identifier The nominal path distance D given by the topology parameter library, the nominal speed v of the conveyor line, and the nominal waiting time given by the process parameter library. and the physical arrival timestamp sent by the programmable logic controller. The output is: position deviation sequence sample and drift detection indicators;

[0127] The nominal motion time is obtained by dividing the nominal distance of the path by the nominal speed, and then adding the nominal waiting time. ;in, Nominal exercise time; The nominal distance of the path from the current segment to the target node is obtained by summing the edge lengths from the topology parameter library; The nominal speed of the conveyor line in this section is read from the equipment parameter library; This is the nominal waiting time, used to describe buffer wait or tick wait, and is read from the process parameter library. If there is no wait, it is set to 0.

[0128] The predicted arrival time stamp is obtained by adding the instruction issuance time stamp to the nominal movement time;

[0129] The arrival deviation is obtained by subtracting the actual arrival time stamp from the predicted arrival time stamp. Let the deviation of the k-th arrival be denoted as... : ;in, This represents the deviation at position k. This is the actual arrival timestamp for the kth arrival. Let k be the predicted arrival timestamp for the kth arrival.

[0130] Will Write deviation sliding window Window length set to Examples are acceptable Maintenance is carried out separately in the welding workshop and the painting workshop;

[0131] For example, the nominal distance of a certain path is D = 18 meters, the nominal speed is v = 0.30 meters per second, and the nominal waiting time is... Seconds, then the nominal motion time If the instruction issuance timestamp is 10:00:00, the predicted arrival timestamp is 10:01:10; if the actual arrival timestamp is 10:01:16, the deviation e = 6 seconds.

[0132] Two-sided cumulative sum and change point detection are used to identify structural changes that are continuously positive or continuously negative in the deviation sequence;

[0133] For window Calculate the absolute deviation of the median from the median: ;in, For the workshop The midpoint of the deviation window; For the workshop The deviation sliding window;

[0134] Calculate the absolute deviation of the deviation window in the workshop: ;e is the window Any biased sample; This is the median value within the deviation window;

[0135] Reference value Used to distinguish between acceptable small random fluctuations and required cumulative deviations, it is defined as the median absolute deviation multiplied by a reference coefficient: ;in, These are reference values ​​for the workshop / shop. For reference coefficients; The absolute deviation is the median of the deviation window.

[0136] For custom coefficients, the following example is available. ,when When setting a lower limit for the reference value. For example, 0.5 seconds to ensure the detection is operational and to set the effective reference value. ;

[0137] Determination threshold Used to determine whether the cumulative amount has reached the change point trigger, defined as the median absolute deviation multiplied by the threshold coefficient: ;in, The threshold for determining the workshop (shop) status; This is the threshold coefficient; The absolute deviation is the median of the deviation window.

[0138] For custom coefficients, the following example is available. ,when When setting a lower threshold For example, 5 seconds to avoid false triggering caused by a threshold of 0;

[0139] Initialize positive accumulator With negative cumulative amount For each new bias sample Recursion: ;in, This is the positive cumulative sum statistic of the workshop at the k-th update; This is the positive cumulative sum statistic from the previous time step; This represents the deviation at position k. For reference only;

[0140] Calculate the negative cumulative sum statistic of the shop at the k-th update: ;in, This is the negative cumulative sum statistic from the previous time step; This represents the deviation at position k. For reference only;

[0141] like or If structural drift is detected, a drift flag is set. And record the trigger time, trigger workshop, and current threshold snapshot;

[0142] For example, if the absolute deviation of the median deviation window in a certain workshop is 2 seconds, then take... Get reference value Seconds, take If the threshold H is 10 seconds, and the deviation is positive for about 4 seconds in a row, the positive cumulative amount will increase by about 3 seconds each time. After about 4 times, it will exceed 10 seconds and trigger drift.

[0143] Drift triggering means that there is a continuous deviation between the physical domain feedback and the prediction model, which may be caused by changes in latency level, increased missed readings, changes in queue congestion mode, or deviation of the conveyor line speed from the nominal value. Without stopping production, two types of updates are performed: one is to make the latency confidence bound of step S1 quickly follow the new latency level; the other is to re-estimate the state model parameters of step S2 so that the logical state reconstruction adapts to the new observation characteristics.

[0144] Update to step S1: Set the process noise amplification factor. Examples are acceptable And amplify the process noise variance of the trigger workshop: ;in, To trigger the current process noise variance in the workshop; This is the process noise amplification factor;

[0145] Increasing the Kalman filter prediction covariance increases the Kalman gain, allowing the filter to absorb new observations more quickly, thus achieving... and Fast convergence, the amplification duration can be configured to... For example, after 50 updates, resume the robust estimation method based on step S1. ;

[0146] The posterior error covariance is reset to a robust scale-squared value based on the bias window to reflect the current increase in uncertainty level: ;in, To determine the posterior error covariance of the trigger workshop filter; To trigger the median absolute deviation of the workshop deviation window;

[0147] Continue recursively applying the prediction and correction formulas from step S1. and ;

[0148] Update to step S2: When And recently Within a time period, for example, if the number of drift triggers exceeds a threshold within 30 minutes. For example, twice, triggering the retraining of the state model in step S2, in order to avoid the fact that emission attenuation alone is insufficient to correct long-term mismatch;

[0149] Extract the set of observation symbol sequences from the most recent M minutes, for example, 60 minutes, from the historical log. Simultaneously, the sequence of physical arrival events within the same time period is extracted for alignment verification. The training input remains the set of observed symbol sequences, and the training output is the updated hidden Markov model state transition probability matrix. With base launch probability ;

[0150] Unsupervised training is performed using the counting normalization method given in step S2. The updated parameters are written back to the state model parameter pool, and the model version number is recorded. With the effective date;

[0151] When drift is triggered, the reliability coefficient of the acquisition terminal can be adjusted based on the statistical changes in the out-of-bounds amount and the degree of conflict. For example, if the average out-of-bounds amount of a certain terminal type increases significantly during drift, a reliability coefficient attenuation factor can be set. Example value: 0.9, and update: ;in, The reliability coefficient for this type of data acquisition terminal; Reliability coefficient and attenuation coefficient;

[0152] The updated reliability coefficient will directly affect the reliability of evidence in step S2. The calculation reduces the impact of unstable terminals on the fusion results;

[0153] By introducing the physical position feedback of the programmable logic controller and constructing the deviation sequence, online drift detection and closed-loop adaptive update of model mismatch are realized: on the one hand, the time delay confidence bound of step S1 can quickly follow the new delay level; on the other hand, the state model parameters of step S2 can be retrained and written back to take effect when needed, thereby ensuring the long-term stable operation of the feasibility determination of step S3 and the automatic scheduling of the routing controller.

[0154] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0155] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0156] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0157] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0158] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0159] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for integrated production and testing line control of automotive modules, characterized in that, Includes steps; Step S1: Construct event objects and calculate observation delays. Perform state estimation for the welding and painting workshops using a Kalman filter. Initialize filter parameters using the observation delay sequence, median, and median absolute deviation. Combine the dynamic estimation of process noise and observation noise to recursively calculate the equivalent delay mean and estimation uncertainty, thereby generating the delay confidence bound for each workshop. Step S2: Calculate the over-boundary quantity of the event based on the time delay confidence bound, construct the time delay attenuation factor, calculate the value by which the event observation delay exceeds the corresponding time delay confidence bound as the over-boundary quantity, and obtain the time delay attenuation factor through exponential mapping based on the over-boundary quantity. The time delay attenuation factor is used to characterize the confidence of the observation value. Construct a hidden Markov model to perform probabilistic inference of the vehicle logic state. Introduce the time delay attenuation factor in the emission probability calculation of the hidden Markov model, so that the contribution of the current observation value to the state estimation decreases as the attenuation factor decreases, realizing dynamic adjustment of the emission probability. In the multi-terminal conflict scenario, introduce evidence theory fusion. For conflict information reported simultaneously by multiple acquisition terminals, determine the basic probability allocation of each terminal's evidence based on the reliability coefficient and time delay attenuation factor of each acquisition terminal. Use the Dempster synthesis rule in evidence theory to fuse multiple pieces of evidence and output the vehicle logic state, confidence level, and conflict degree. Step S3: Using the confidence and conflict history window, construct an adaptive threshold by the median and the absolute deviation of the median to determine the feasibility of the vehicle status. If the confidence is higher than the lower confidence threshold and the conflict is lower than the upper conflict threshold, the scheduling is determined to be feasible. The vehicle status is input into the routing controller to perform path calculation and issue control commands. Otherwise, the scheduling is determined to be infeasible. Abnormal diversion processing is performed and alarm information is pushed. Step S4: Construct an arrival deviation sequence based on the physical arrival timestamp fed back by the programmable logic controller and the predicted arrival timestamp. For each workstation, record the difference between the physical arrival timestamp fed back by the programmable logic controller and the predicted arrival timestamp to form an arrival deviation time sequence. Detect structural drift using the cumulative sum method. When structural drift is detected, increase the process noise variance of the Kalman filter and reset the posterior error covariance. When the number of structural drift triggers within a preset time window exceeds a preset threshold, retrain the hidden Markov model parameters using the data from the current time period and adjust the reliability coefficient of the acquisition terminal according to the degree of drift.

2. The integrated production and testing assembly line control method for automotive modules according to claim 1, characterized in that, The median and median absolute deviation are calculated based on historical observation time delay sequences. The median is used as the initial state estimate of the Kalman filter. The median absolute deviation is converted into a robust scale and then squared as the initial covariance estimate of the Kalman filter.

3. The integrated production and testing assembly line control method for automotive modules according to claim 2, characterized in that, During the recursive process of the Kalman filter, the equivalent delay mean and estimation uncertainty are corrected in real time based on the statistical characteristics of the current observation delay update process noise and observation noise, according to the updated statistical characteristics.

4. The integrated production and testing assembly line control method for automotive modules according to claim 1, characterized in that, The confidence level and conflict level of the most recent preset number of time moments are collected to form historical windows. The median and absolute deviation of the median of the data within the historical confidence window are calculated to generate the lower confidence threshold. The median and absolute deviation of the median of the data within the historical conflict level are calculated to generate the upper conflict level threshold.

Images