A quality target-oriented manufacturing process optimization decision method and system

Abstract

The application relates to the technical field of intelligent manufacturing process control, and discloses a manufacturing process optimization decision method and system for a quality target. The method acquires controllable process parameters, sensing time sequence data, a quality target, a safety boundary and available recovery resources, constructs a time sequence causal network containing candidate causal propagation relations, generates a zero net increment pairing check sequence, forms an expected check result and a recoverable state domain, and determines a target adjustment unit; according to a measured check result, execution basis is determined, candidate execution increments are applied step by step, and quality improvement or execution recovery is confirmed under the constraints of safety and recovery resources. The application solves the problems that the existing method is unreliable in adjustment basis and difficult to control in execution risk when process relations are uncertain, and improves the accuracy, safety and recoverability of quality optimization decision.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing process control technology, and in particular to a manufacturing process optimization decision-making method and system oriented towards quality objectives. Background Technology

[0002] During the manufacturing process, product quality is typically influenced by a combination of factors, including process parameters, equipment operating status, raw material characteristics, and the production environment. To ensure that product quality meets predetermined targets, existing technologies usually collect operational and quality inspection data during the manufacturing process, establish a correlation between quality indicators and process parameters, and generate corresponding process parameter adjustment plans based on quality deviations.

[0003] Because manufacturing processes are dynamic, coupled, and lag-dependent, the actual influence relationship between process parameters and quality indicators may change with equipment status, production batches, and operating conditions. Existing manufacturing process optimization methods typically determine adjustment schemes directly based on historical data, current test results, or established models. It is difficult to fully determine whether the process relationships on which the methods are based are consistent with the current operating conditions before the adjustment is implemented. Therefore, the resulting adjustment schemes may not be consistent with actual quality changes.

[0004] If the basis for adjustment is inaccurate, adjusting process parameters may fail to effectively reduce quality deviations and could lead to process fluctuations, equipment deviating from normal operating ranges, or further deterioration of product quality. Finding suitable parameters through repeated trial and error can also increase production interruptions, raw material losses, and equipment operational risks. Therefore, improving the reliability of manufacturing process optimization decisions and reducing the process risks caused by erroneous adjustments while implementing quality adjustments is a problem that existing manufacturing process optimization technologies need to address. Summary of the Invention

[0005] The purpose of this invention is to provide a manufacturing process optimization decision-making method and system oriented towards quality objectives, in order to solve the problem that existing manufacturing process optimization methods are unable to accurately determine the effective process relationship corresponding to quality deviations under dynamic changes in operating conditions and multi-factor coupling conditions, resulting in insufficient reliability of process parameter adjustment and difficulty in controlling process risks, thereby improving the accuracy, safety, recoverability and adaptability of manufacturing process quality optimization decisions.

[0006] To achieve the above objectives, the present invention employs the following technical solution:

[0007] This invention provides a manufacturing process optimization decision-making method oriented towards quality objectives, comprising the following steps:

[0008] S1. Obtain controllable process parameters, sensing timing data, quality targets, safety boundaries, and available recovery resources to determine quality deviations; construct a time-series causal network, set multiple candidate causal propagation relationships and time delay ranges and response envelopes to form the current causal network version and the safe and maintainable state domain;

[0009] S2. Generate candidate execution increments. Generate a pairing verification sequence consisting of positive perturbation and anti-polarity reverse time-series perturbation with an increment algebraic sum of zero in the controllable process parameter channel. Predict the pairing response of each candidate causal propagation relationship. After time delay alignment and pairing decomposition, form the expected verification result containing causal response components and pairing closure residuals.

[0010] S3. Based on the recovery requirements and available recovery resources when the pairing verification is aborted and the candidate execution increment fails, a recoverable state domain is formed. Candidate execution increments whose verification state envelope and execution state envelope are located in the recoverable state domain, whose recovery state envelope is located in the safe and maintainable state domain, and whose manufacturing process can be restored to the preset safe process state are retained.

[0011] S4. Combine the retained candidate execution increments with the corresponding paired verification sequences to form a two-stage adjustment unit. Determine the target adjustment unit based on the distinguishability of the candidate causal propagation relationship and the amount of reduction in prediction quality deviation.

[0012] S5. Apply the pairing verification sequence of the target adjustment unit and match the measured verification result with the expected verification result. When there is a unique match, apply the candidate execution increment step by step based on the corresponding candidate causal propagation relationship. When the recoverable state domain and recovery resource constraints are met during the step-by-step application, and the quality deviation is reduced within the corresponding time delay range after the application is completed, the candidate execution increment is determined as the formal process adjustment amount and the application is maintained. If any condition is not met, stop the execution and restore to the preset safe process state.

[0013] Furthermore, the current causal network version and the secure, maintainable state domain are formed in the following manner:

[0014] Controllable process parameters, process states represented by sensing time-series data, and quality indicators corresponding to quality targets are set as network nodes. Directed propagation edges are established according to the direction of influence transmission. Multiple candidate causal propagation relationships are generated for each directed propagation edge according to process constraints. The time delay range and response envelope of each candidate causal propagation relationship are set.

[0015] The initial state set is the set of states that satisfy the safety boundary and equipment constraints. For each state to be screened, the next state envelope corresponding to each controllable process parameter increment that satisfies the equipment constraints is predicted by using the time delay range and response envelope of each candidate causal propagation relationship. When there is a controllable process parameter increment that makes the next state envelope corresponding to all candidate causal propagation relationships within the state set of this round, the state to be screened is retained. The state is iterated until no more states are eliminated in this round, and the retained state set is determined as the safe and maintainable state domain.

[0016] Associating the temporal causal network, candidate causal propagation relationships, corresponding time delay ranges and response envelopes, and secure and maintainable state domains with the same version identifier forms the current causal network version.

[0017] Further, the process of generating the pairwise verification sequence and predicting the paired responses of each candidate causal propagation relationship includes:

[0018] A forward perturbation sequence is generated in the same controllable process parameter channel. The perturbation quantities in the forward perturbation sequence are inverted in the opposite time sequence to form an anti-polarity inverse time sequence perturbation sequence, so that the algebraic sum of the controllable process parameter increments corresponding to the forward perturbation sequence and the anti-polarity inverse time sequence perturbation sequence is zero.

[0019] For each candidate causal propagation relationship in the current causal network version, based on the time delay range and response envelope, the sensor responses caused by the forward perturbation sequence and the reverse polarity reverse time-series perturbation sequence are predicted respectively. The predicted baseline response of the same manufacturing cycle when no corresponding perturbation sequence is applied is subtracted respectively. The predicted responses corresponding to the reverse polarity reverse time-series perturbation are rearranged in reverse time order without changing the response polarity, and the predicted forward response increment sequence and the predicted reverse polarity response increment sequence are formed according to the same sensor channel order.

[0020] The forward perturbation sequence, the reverse polarity inverse time-series perturbation sequence, and the state envelopes corresponding to the connection process of the two sequences are used to form the verification state envelope. Paired verification sequences whose verification state envelopes are located in the safe and maintainable state domain and whose quality perturbations do not exceed the verification tolerance are retained. The retained paired verification sequences are associated with the corresponding candidate execution increments.

[0021] Furthermore, the expected verification result is formed through the following steps:

[0022] The time delay range of each candidate causal propagation relationship is discretized according to the sampling period of the sensing time series data; for each discrete time delay, a set of effective sampling indices with corresponding sample values ​​for both the predicted positive response increment sequence and the predicted reverse polarity response increment sequence is determined; discrete time delays with effective sampling numbers reaching a preset minimum effective sampling number are combined into a set of effective candidate aligned time delays, and a pairwise verification sequence in which the effective candidate aligned time delay sets of each candidate causal propagation relationship are not empty is retained; according to the same sensing channel order, an invertible diagonal response scale matrix is ​​constructed based on the response scales of each sensing channel that are greater than zero;

[0023] Candidate causal propagation relationships are calculated using the following formula. Candidate alignment delay The following are the pairing closure energy ratios:

[0024] ;

[0025] In the formula, As a candidate causal propagation relationship identifier, The candidate alignment delay is expressed in terms of the number of sampling beats. Candidate alignment delay The corresponding set of valid sampling indices, To determine the effective number of samples, For effective sampling index, For the first The predicted positive response increment vector for each sampling tick. For reverse time rearrangement and candidate alignment delay Aligned predicted reverse polarity response increment vectors The response scaling matrix has diagonal elements that have the same dimensions as the corresponding sensing response. Assuming a dimensionless positive number, The ratio of paired closure energies. Represents the 2-norm of a vector;

[0026] The selected alignment delay is determined from the candidate alignment delays with the smallest pairing closure energy ratio. Under the selected alignment delay, half of the difference between the predicted positive response increment vector corresponding to each effective sampling index and the aligned predicted reverse polarity response increment vector is used to form the causal response component. Half of the sum of the two is used to form the pairing closure residual. The selected alignment delay, the corresponding pairing closure energy ratio, the causal response component, and the pairing closure residual are used to form the expected verification result of the corresponding candidate causal propagation relationship.

[0027] Further, forming the recoverable state domain and retaining the candidate execution increments includes:

[0028] The states that may be reached when the pairing verification sequence is terminated and the states that may be reached when the candidate execution incremental step-by-step application fails are determined as the recovery start state;

[0029] For each recovery start state, determine the same recovery increment sequence that satisfies the device constraints. Based on the time delay range and response envelope of all candidate causal propagation relationships, predict the recovery state envelope and recovery requirements of various recovery resources corresponding to the same recovery increment sequence. Determine resource reservation requirements based on the upper bound of the recovery requirements of various recovery resources under each candidate causal propagation relationship.

[0030] Under all candidate causal propagation relationships, the recovery state envelope remains in the safe and maintainable state domain. The recovery increment sequence enables the manufacturing process to enter a preset safe process state. When the available recovery resources meet the resource reservation requirements, the corresponding recovery start state is included in the recoverable state domain.

[0031] Candidate execution increments whose verification state envelope and execution state envelope are both located in the recoverable state domain are retained.

[0032] Furthermore, the target adjustment unit is determined according to the following rules:

[0033] The retained candidate execution increments are combined with the corresponding paired verification sequences to form a two-stage adjustment unit set;

[0034] According to the preset resolution scale corresponding to each component of the expected verification result, the expected verification result corresponding to each candidate causal propagation relationship under each two-stage adjustment unit is quantified, and the candidate causal propagation relationships with the same quantification result are divided into the same indistinguishable relationship group.

[0035] The number of candidate causal propagation relations contained in the indistinguishable relation group with the largest number of members is determined as the maximum size of the indistinguishable relation group corresponding to the two-stage adjustment unit, and the amount of reduction in prediction quality deviation is determined based on the prediction quality deviation before and after the application of the candidate execution increment.

[0036] Select the two-stage adjustment unit with the smallest maximum size of the indistinguishable relation group; when there are multiple two-stage adjustment units with the same maximum size of indistinguishable relation groups, select the two-stage adjustment unit with the largest reduction in prediction quality deviation as the target adjustment unit.

[0037] Furthermore, the process of applying the pairing verification sequence to the target adjustment unit includes:

[0038] Establish execution association relationships between the target adjustment unit and the current causal network version, the corresponding recovery increment sequence, and the available recovery resources that meet the resource reservation requirements, and lock the available recovery resources corresponding to the execution association relationships during the pairing verification sequence and the application of candidate execution increments;

[0039] According to the manufacturing rhythm, the forward perturbation sequence and the reverse polarity inverse time perturbation sequence are applied sequentially. The measured response of each sensing channel is collected. During the application of the pairing verification sequence, the verification state envelope is predicted based on the candidate causal propagation relationship in the current causal network version. When the verification state envelope exceeds the recoverable state domain or the available recovery resource margin does not meet the resource reservation requirements, the pairing verification sequence is stopped, and the recovery increment sequence corresponding to the execution association relationship is applied.

[0040] After the paired verification sequence is applied, the predicted baseline response of the same manufacturing cycle when no corresponding perturbation sequence is applied is subtracted. The measured responses corresponding to the reverse polarity reverse time-series perturbation are rearranged in reverse time sequence without changing the response polarity to form the measured forward response increment sequence and the measured reverse-order reverse polarity response increment sequence. According to the time delay alignment and paired decomposition method used when forming the expected verification results, the measured causal response components and measured paired closure residuals corresponding to each candidate causal propagation relationship are formed, and the measured verification results are composed.

[0041] Furthermore, based on the measured verification results, the execution criteria are determined and the candidate execution increments are applied step by step:

[0042] According to the preset resolution scale, the measured verification results are quantitatively matched with the expected verification results corresponding to each candidate causal propagation relationship; when only one candidate causal propagation relationship satisfies the same quantification result and the difference between each component does not exceed the corresponding preset resolution scale, the corresponding candidate causal propagation relationship is determined as the execution basis; when a unique match cannot be made, no candidate execution increment is applied, and a recovery increment sequence corresponding to the execution association relationship is applied.

[0043] When the current causal network version, target adjustment unit, recovery increment sequence, and locked available recovery resources associated with the execution relationship remain unchanged, candidate execution increments are applied step by step according to the manufacturing cycle, and the execution state envelope is predicted based on the execution basis and the current process state; during the step-by-step application, if the execution state envelope exceeds the recoverable state domain or the available recovery resource margin does not meet the resource reservation requirements, the application of the remaining part of the candidate execution increments is stopped, and the recovery increment sequence is applied.

[0044] After the candidate execution increment is applied in steps, if a decrease in quality deviation is detected within the time delay range corresponding to the execution basis compared to before the application, the candidate execution increment is determined as the formal process adjustment amount and is maintained; if no decrease in quality deviation is detected, the recovery increment sequence is applied.

[0045] Furthermore, the current causal network version is updated based on anomalous observations in the following manner:

[0046] Anomalies are determined based on the deviations between measured causal response components and measured paired closure residuals and corresponding expected components.

[0047] When the deviation of the causal response component exceeds the corresponding preset resolution scale and the deviation of the paired closure residual does not exceed the corresponding preset resolution scale, the abnormal observation is identified as a change in the candidate causal propagation relationship; when the deviation of the paired closure residual exceeds the corresponding preset resolution scale, the abnormal observation is identified as an anomaly in the execution measurement chain.

[0048] In response to changes in candidate causal propagation relationships, the time delay range or response envelope of the corresponding candidate causal propagation relationships is modified, and a safe and maintainable state domain and a recoverable state domain are re-formed. The current causal network version is updated when both the safe and maintainable state domain and the recoverable state domain are not empty.

[0049] When an abnormal observation is identified as an execution measurement chain anomaly, or when the newly formed safe and maintainable state domain or recoverable state domain is empty, the controllable process parameters will be switched to the preset safe process parameter group.

[0050] The present invention also provides a manufacturing process optimization decision system oriented towards quality objectives, for executing the manufacturing process optimization decision method oriented towards quality objectives as described above, comprising:

[0051] The data modeling module is used to acquire controllable process parameters, sensor time series data, quality targets, safety boundaries and available recovery resources, determine quality deviations, construct a time series causal network, set multiple candidate causal propagation relationships and their time delay range and response envelope, and form the current causal network version and the safe and maintainable state domain.

[0052] The verification generation module is used to generate candidate execution increments. It generates a paired verification sequence consisting of positive perturbations and anti-polarity inverse time-series perturbations with an increment algebraic sum of zero in the controllable process parameter channel. It predicts the paired response of each candidate causal propagation relationship and forms the expected verification result containing causal response components and paired closure residuals after time delay alignment and paired decomposition.

[0053] The recovery filtering module is used to form a recoverable state domain based on the recovery requirements and available recovery resources when the pairing verification is aborted and the candidate execution increment fails. It retains candidate execution increments whose verification state envelope and execution state envelope are located in the recoverable state domain, whose recovery state envelope is located in the safe and maintainable state domain, and whose manufacturing process can be restored to the preset safe process state.

[0054] The adjustment unit determination module is used to form a two-stage adjustment unit by combining the retained candidate execution increments with the corresponding paired verification sequences, and to determine the target adjustment unit based on the distinguishability of the candidate causal propagation relationship and the amount of reduction in prediction quality deviation.

[0055] The execution control module is used to apply the pairing verification sequence of the target adjustment unit and match the measured verification result with the expected verification result. When there is a unique match, the candidate execution increment is applied step by step based on the corresponding candidate causal propagation relationship. When the recoverable state domain and recovery resource constraints are met during the step-by-step application, and the quality deviation is reduced within the corresponding time delay range after the application is completed, the candidate execution increment is determined as the formal process adjustment amount and the application is maintained. If any condition is not met, the execution is stopped and the manufacturing process is restored to the preset safe process state.

[0056] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention constructs a temporal causal network containing multiple candidate causal propagation relationships. It generates a paired verification sequence consisting of forward perturbations and anti-polarity inverse temporal perturbations with an incremental algebraic sum of zero in the controllable process parameter channel. Causal response components and paired closure residuals are obtained through time delay alignment and paired decomposition. Furthermore, the target adjustment unit is selected by combining the distinguishability of candidate causal propagation relationships and the predicted reduction in quality deviation. The execution basis is determined after a unique match in the measured verification results. This invention can distinguish between true causal responses and non-closed abnormal responses before formally implementing process adjustments, reducing the risk of erroneous adjustments caused by misjudgments of causal relationships, sensor disturbances, or transient fluctuations, and improving the accuracy and interpretability of manufacturing process quality optimization decisions.

[0057] This invention further establishes a safe and maintainable state domain and a recoverable state domain based on safety boundaries, the response envelope of candidate causal propagation relationships, and available recovery resources. Before applying candidate execution increments, the safety of the verification state, execution state, and recovery state is verified, and the state envelope and recovery resources are monitored step-by-step during execution. Only after confirming a reduction in quality deviation within the corresponding time delay range is the candidate execution increment determined as a formal process adjustment; otherwise, execution is stopped and the process reverts to the preset safe process state. Thus, quality improvement judgment, process safety constraints, and failure recovery capabilities are uniformly incorporated into the optimization decision-making closed loop, improving the reliability of achieving quality objectives while avoiding irreversible deviations or safety risks caused by process adjustments. Attached Figure Description

[0058] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the pairing verification sequence generation and time delay alignment decomposition process in an embodiment of the present invention; Figure 3This is a schematic diagram of the process for forming a recoverable state domain and incrementally filtering candidates for execution in an embodiment of the present invention; Figure 4 This is a schematic diagram of the target adjustment unit execution and safety closed-loop control process in an embodiment of the present invention; Figure 5 This is a system structure block diagram according to an embodiment of the present invention. Detailed Implementation

[0059] The following describes the implementation of this invention in conjunction with a round of manufacturing process optimization decision-making. This embodiment is applicable to manufacturing processes where changes in equipment status, material batches, and operating conditions prevent the causal propagation relationship from being uniquely determined in advance. By performing a low-net-value pairing verification before formally applying process adjustments, the actual execution basis is determined under the condition that the entire process has safe recovery capabilities, and the formal process adjustment amount is confirmed based on the quality improvement results.

[0060] like Figure 1 As shown, this is an embodiment of the present invention, which provides a manufacturing process optimization decision-making method oriented towards quality objectives, including the following steps:

[0061] S1. Obtain controllable process parameters, sensing timing data, quality targets, safety boundaries, and available recovery resources to determine quality deviations; construct a time-series causal network, set multiple candidate causal propagation relationships and time delay ranges and response envelopes to form the current causal network version and the safe and maintainable state domain;

[0062] The controllable process parameters are process variables that can be adjusted by the manufacturing control system or field actuators, including temperature setpoints, pressure setpoints, flow rates, processing speeds, and valve openings. For each controllable process parameter, the current value, allowable range, maximum adjustment amount per instance, rate of change, and execution cycle are obtained to form corresponding equipment constraints.

[0063] The sensing time-series data comprises process status data and quality-related data acquired according to the sampling cycle or manufacturing cycle, including temperature, pressure, flow rate, load, material state, and quality inspection results. Each sensing time-series data is associated with a sensing channel identifier, sampling time, and sampled value. Time alignment, outlier identification, and missing value processing are performed on the data from different sensing channels to match the sampled values ​​of each sensing channel with the corresponding manufacturing cycle. Data that remains invalid after processing is not used for establishing candidate causal propagation relationships or calculating verification results.

[0064] The quality target is the desired quality index value or quality index range to be achieved in the manufacturing process. The quality deviation is determined based on the difference between the current quality index and the corresponding target value or the boundary of the adjacent target range.

[0065] When the quality objective includes multiple quality indicators, the quality deviations are grouped into a quality deviation vector according to a preset quality indicator order, and then scaled according to the corresponding allowable tolerances to form a dimensionless quality deviation vector. The quality deviation values ​​are then determined based on the dimensionless quality deviation vector according to a preset quality deviation aggregation rule. The quality deviation aggregation rule can be the sum of the absolute values ​​of the dimensionless quality deviations, the maximum value, or a weighted sum. The same quality indicator order, scale conversion method, and quality deviation aggregation rule are used in the same round of manufacturing process optimization decisions.

[0066] The safety boundaries are used to define the safe range that the manufacturing process state must not exceed, including upper limits for process temperature, process pressure, equipment load, and quality index boundaries that may cause irreversible quality damage. Safety boundaries are used to determine whether the manufacturing process state is safe, while equipment constraints are used to limit whether control actions can be actually executed by the manufacturing equipment.

[0067] The available recovery resources are those used to restore the manufacturing process to a preset safe process state when pairing verification is aborted or candidate execution increment fails. These resources include standby adjustment capacity, cooling capacity, heating capacity, buffer capacity, and standby power. For each type of available recovery resource, the current available quantity, maximum output rate, response delay, and duration are obtained to form an available recovery resource record.

[0068] After obtaining the above data, a temporal causal network is constructed. This network includes network nodes, directed propagation edges, and candidate causal propagation relationships associated with these edges. Controllable process parameters are set as control-type network nodes, process states represented by sensing time-series data are set as state-type network nodes, and quality indicators are set as quality-type network nodes. Each network node is associated with its data source, current value, sampling period, physical dimensions, and allowable range.

[0069] Based on the material flow, energy transfer path, equipment connection relationship, process sequence and feedback control relationship in the manufacturing process, determine the allowable influence transmission direction between network nodes and establish directed propagation edges. Based on process constraints, eliminate influence directions and node connection relationships that do not conform to the actual manufacturing mechanism.

[0070] Due to changes in equipment status, material batches, and operating conditions, the propagation delay, response amplitude, or response direction corresponding to the same directed propagation edge may be uncertain. Therefore, multiple candidate causal propagation relationships are set for each directed propagation edge. Each candidate causal propagation relationship is associated with the source network node, the target network node, the direction of action, the delay range, and the response envelope.

[0071] The time delay range represents the time range within which the target network node may respond after a change in the source network node. It is determined based on the equipment response time, material conveying time, process reaction time, and historical sensor timing data, and is expressed as actual time or sampling cycle number.

[0072] The response envelope represents the range of possible responses from the target network node when a given change occurs in the source network node within a corresponding time delay range. It is determined based on the process mechanism, equipment technical parameters, and historical response data, and covers the uncertainties associated with equipment errors, measurement errors, and operating condition fluctuations. When the response envelope is represented in interval form, representative response values ​​are determined from the response envelope according to the representative response value determination rules associated with the current causal network version; the same representative response value determination rules are used for the same current causal network version.

[0073] After forming a temporal causal network, the manufacturing process states that satisfy the safety boundary constitute an initial state set, and equipment constraints limit the controllable process parameter increments that each state to be screened can adopt. These controllable process parameter increments are used to determine whether the manufacturing process state possesses continuous safety and controllability capabilities, and are distinct from the candidate execution increments used for quality adjustment in step S2.

[0074] For each state to be screened in the initial state set and each controllable process parameter increment that satisfies the equipment constraints, the change amount of the corresponding control class network node is determined according to the controllable process parameter increment, and the manufacturing process state change is predicted along the directed propagation edge.

[0075] For each candidate causal propagation relationship, the response interval of the target network node within the corresponding time delay range is determined based on the change amount and response envelope of the source network node; the response interval is combined with the baseline state range when no controllable process parameter increment is applied to obtain the predicted state range of the target network node; the above processing is performed step by step according to the influence transmission order of the directed propagation edge, and the predicted state ranges of each state-type network node and quality-type network node are combined into the next state envelope under the corresponding candidate causal propagation relationship.

[0076] When the same target network node is affected by multiple upstream network nodes, the response intervals corresponding to each upstream network node are combined according to the preset process coupling relationship. For responses with superimposed relationships, the execution intervals are superimposed, and for responses that are not suitable for direct superposition, a joint envelope covering all possible response combinations is formed.

[0077] For each state to be screened, determine whether there exists a controllable process parameter increment that satisfies the equipment constraints, such that the envelope of the next state corresponding to all candidate causal propagation relationships is within the current state set. If the controllable process parameter increment exists, retain the corresponding state to be screened; otherwise, discard it.

[0078] The above filtering process is repeated on the retained state set until no more states are eliminated in a complete filtering process. The final retained state set is then determined as the safe and maintainable state domain. The safe and maintainable state domain means that for any manufacturing process state within the domain, there exists a controllable process parameter increment that satisfies the equipment constraints, such that the next state envelope corresponding to all candidate causal propagation relationships remains within the domain.

[0079] Associating the temporal causal network, candidate causal propagation relationships, time delay range, response envelope, secure and maintainable state domain, and representative response value determination rules with the same version identifier forms the current causal network version. Step S1 outputs the quality deviation, quality deviation value, current causal network version, secure and maintainable state domain, and available recoverable resource records.

[0080] S2. Generate candidate execution increments. Generate a pairing verification sequence consisting of positive perturbation and anti-polarity reverse time-series perturbation with an increment algebraic sum of zero in the controllable process parameter channel. Predict the pairing response of each candidate causal propagation relationship. After time delay alignment and pairing decomposition, form the expected verification result containing causal response components and pairing closure residuals.

[0081] Multiple candidate execution increments are generated based on quality deviation, current causal network version, and equipment constraints. Specifically, based on the candidate causal propagation relationship between each controllable process parameter channel and the quality index where the quality deviation occurs, controllable process parameter channels and adjustment directions that may reduce quality deviation are determined. Within the allowable value range, maximum single adjustment amount, and rate of change range of the corresponding controllable process parameter, the executable adjustment amplitude is discretized or combined to form candidate execution increments.

[0082] Each candidate execution increment includes a controllable process parameter channel, adjustment direction, adjustment magnitude, and step-by-step application method. The adjustment magnitude of the candidate execution increment does not exceed the maximum single adjustment allowed by equipment constraints, and the step-by-step application method is determined based on the manufacturing cycle time and the rate of change of the corresponding controllable process parameter.

[0083] For each candidate execution increment, a set of verification response channels is selected from the current causal network version. These channels are influenced by the corresponding controllable process parameter channels, have valid sensor data, and can be used to distinguish candidate causal propagation relationships. The expected verification results and the measured verification results use the same set of verification response channels and the same sensor channel order.

[0084] A paired verification sequence is generated in the controllable process parameter channel corresponding to the candidate execution increment. The paired verification sequence includes a positive perturbation sequence and an anti-polarity reverse-time perturbation sequence. The positive perturbation represents a perturbation sequence applied in the initial generation order, and it is not required that each perturbation quantity is positive.

[0085] like Figure 2 As shown, the paired verification sequence is composed of a forward perturbation sequence and an anti-polarity inverse time-series perturbation sequence. After the response is acquired, baseline subtraction, time delay alignment and paired decomposition are performed to obtain the expected verification results of the corresponding candidate causal propagation relationship.

[0086] The positive perturbation sequence consists of multiple perturbations arranged according to the manufacturing cycle. The absolute value of each perturbation is greater than the minimum executable increment of the corresponding controllable process parameter channel, and the effective predicted response in the verification response channel set reaches the corresponding identifiable response lower limit; the absolute value of each perturbation does not exceed the single adjustment amount allowed by the equipment constraints.

[0087] The minimum executable increment is determined based on the control resolution, dead zone, and minimum stable output capability of the actuator; the identifiable response lower limit is determined based on the sensor resolution, normal operating condition fluctuation range, and measurement noise level; the perturbation sequence length and perturbation duration beat count are determined based on the device response period, sampling period, and time delay range of candidate causal propagation relationships.

[0088] Arrange the perturbations in the forward perturbation sequence in reverse temporal order and then invert them to form an inverse polarity reverse temporal perturbation sequence. For example, the forward perturbation sequence is as follows: , and At that time, the reverse polarity inverse time-series perturbation sequences are as follows: , and .

[0089] The incremental algebraic sum of the controllable process parameters corresponding to the forward perturbation sequence and the reverse polarity inverse time-series perturbation sequence is zero. The zero incremental algebraic sum indicates that the paired verification sequence does not form a continuously accumulating parameter offset at the control input level, but does not mean that the manufacturing process state will necessarily be completely restored to the state before verification.

[0090] For each candidate causal propagation relationship in the current causal network version, based on the corresponding time delay range, response envelope, and representative response value determination rules, the sensing responses caused by the forward perturbation sequence and the reverse polarity inverse time perturbation sequence are predicted respectively.

[0091] The perturbation of each manufacturing cycle is input into the corresponding source network node. The predicted response of the verification response channel within the corresponding time delay range is determined step by step along the temporal causal network. The predicted sensing response vector is formed according to the order of the sensing channels in the set of verification response channels. The upper and lower bounds of the response envelope are used to form the verification state envelope.

[0092] Determine the predicted baseline response for the same manufacturing cycle when no positive perturbation sequence and the reverse polarity inverse time-series perturbation sequence are applied, respectively. Subtract the predicted sensor response from the corresponding predicted baseline response to form the predicted positive response increment sequence and the predicted reverse polarity response increment sequence, respectively.

[0093] The predicted reverse polarity response increment sequence is rearranged in reverse time order without changing the response polarity to form the predicted reverse polarity response increment sequence. The predicted positive response increment vector and the predicted reverse polarity response increment vector are then formed according to the same sensing channel order.

[0094] Based on the time delay range and response envelope of each candidate causal propagation relationship, the possible state ranges that the manufacturing process may reach during the forward perturbation sequence, the reverse polarity inverse time-series perturbation sequence, and the connection between the two sequences are predicted, and a verification state envelope is formed. When the verification state envelope is located in the safe and maintainable state domain and the predicted quality perturbations of each quality index do not exceed the corresponding verification tolerance, the paired verification sequence is retained and associated with the corresponding candidate execution increment. The verification tolerance is determined based on the allowable tolerance of the quality index, the quality detection resolution, and the allowable transient quality fluctuation range.

[0095] For the retained pairwise verification sequences, the time delay range of each candidate causal propagation relationship is discretized according to the sampling period of the sensing time series data to form a candidate alignment time delay. For each candidate alignment time delay, a set of valid sampling indices is determined in which both the predicted forward response increment sequence and the aligned predicted reverse polarity response increment sequence have sampled values.

[0096] When the number of valid samples reaches the preset minimum number of valid samples, the corresponding candidate alignment delay is included in the set of valid candidate alignment delays. The preset minimum number of valid samples is determined based on the prediction response duration, sampling period, and computational stability requirements. Only paired verification sequences where the set of valid candidate alignment delays for each candidate causal propagation relationship is not empty are retained.

[0097] Based on the sensor channel order in the verification response channel set, an invertible diagonal response scale matrix is ​​constructed according to the positive-zero response scales of each sensor channel. Each diagonal element of the response scale matrix has the same dimensions as the corresponding sensing response and is determined based on the sensor resolution, the normal operating condition fluctuation scale, or the allowable response error.

[0098] Candidate causal propagation relationships are calculated using the following formula. Candidate alignment delay The following are the pairing closure energy ratios:

[0099] ;

[0100] In the formula, As a candidate causal propagation relationship identifier, The candidate alignment delay is expressed in terms of the number of sampling beats. Candidate alignment delay The corresponding set of valid sampling indices, To determine the effective number of samples, For effective sampling index, For the first The predicted positive response increment vector for each sampling tick. For reverse time rearrangement and candidate alignment delay Aligned predicted reverse polarity response increment vectors The response scaling matrix has diagonal elements that have the same dimensions as the corresponding sensing response. Assuming a dimensionless positive number, The ratio of paired closure energies. Represents the 2-norm of a vector;

[0101] through The transformed response vector is a dimensionless vector; therefore, the paired closed-loop energy ratio is a dimensionless quantity. The calculation is determined based on the computational accuracy and historical normalized response energy, and is used to avoid abnormal amplification of the calculation results when the denominator is zero or the response energy is small.

[0102] The selected alignment delay is determined from the candidate alignment delays with the smallest paired closure energy ratio. Under the selected alignment delay, half of the difference between the predicted positive response increment vector and the aligned predicted reverse polarity response increment vector is used to form the causal response component, and half of the sum of the two is used to form the paired closure residual.

[0103] The causal response component is used to characterize the principal paired response corresponding to the candidate causal propagation relation; the paired closure residual is used to characterize the expected non-closed portion that is not canceled by the principal paired response of the opposite polarity under the current candidate causal propagation relation and its response envelope. The causal response component and the paired closure residual are calculated using the response increment vector without response scaling and have the same dimensions as the corresponding sensing response.

[0104] The selected alignment delay, the corresponding pairing closure energy ratio, the causal response component, and the pairing closure residual are used to form the expected verification result of the corresponding candidate causal propagation relationship.

[0105] Step S2 outputs the candidate execution increment, the paired verification sequence, the verification state envelope, and the expected verification results corresponding to each candidate causal propagation relationship.

[0106] S3. Based on the recovery requirements and available recovery resources when the pairing verification is aborted and the candidate execution increment fails, a recoverable state domain is formed. Candidate execution increments whose verification state envelope and execution state envelope are located in the recoverable state domain, whose recovery state envelope is located in the safe and maintainable state domain, and whose manufacturing process can be restored to the preset safe process state are retained.

[0107] Based on the allowed abort timeframes of the paired verification sequence and the step-by-step execution timeframes of the candidate execution increments, and according to the applied perturbation or execution increment, the current process state, and the time delay range and response envelope of all candidate causal propagation relationships, the possible state range that may be reached after abort or execution failure is predicted, and the state in the state range is determined as the recovery start state.

[0108] For each recovery start state, based on the state deviation between the recovery start state and the preset safe process state, the controllable process parameter channels and adjustment directions for recovery are determined, and a recovery increment sequence is generated within the equipment constraints. The preset safe process state is located within the safe and maintainable state domain and corresponds to the preset safe process parameter set that the manufacturing equipment can execute.

[0109] For a given recovery starting state, the same recovery increment sequence is used to verify the recovery process under all candidate causal propagation relationships. The same recovery increment sequence means that, when the actual candidate causal propagation relationships are not yet uniquely determined, a set of identical recovery increment sequences is used to ensure safe recovery under all candidate causal propagation relationships.

[0110] When the recovery starting state is represented by a state range, the same recovery increment sequence should also be applicable to all possible states within that state range.

[0111] Using the initial recovery state as the initial state, the corresponding controllable process parameters are changed sequentially according to the recovery increment sequence. The manufacturing process state is predicted step-by-step based on the time delay range and response envelope of each candidate causal propagation relationship. The recovery state envelope is formed by the state range formed during the application of the recovery increment sequence and the state range that continues to propagate within the corresponding time delay range after the application is completed.

[0112] Based on the magnitude, duration, actuator efficiency, and resource consumption of the recovery increment, the recovery requirements for each type of recovery resource are determined. For each type of recovery resource, the upper bound of the corresponding recovery requirement under all candidate causal propagation relationships is taken to form the resource reservation requirements.

[0113] The resource reservation requirements are based on the usage characteristics of the restored resources, including the reservation amount, minimum output rate, maximum allowable response delay, and minimum duration; restored resources of different dimensions are calculated separately and are not directly added together.

[0114] Under all candidate causal propagation relationships, the recovery state envelope is located in the safe and maintainable state domain, the recovery increment sequence can enable the manufacturing process to enter the preset safe process state, and when the available recovery resources meet the resource reservation requirements, the corresponding recovery start state is included in the recoverable state domain.

[0115] like Figure 3As shown, a recoverable state domain is formed by performing recovery verification on the pairing verification abort state and the candidate execution increment execution failure state, and candidate execution increments are selected according to the positional relationship between the verification state envelope and the execution state envelope.

[0116] The recoverable state domain means that for any recovery start state within the state domain, there exists a set of recovery increment sequences that satisfy the equipment constraints, enabling the manufacturing process to safely enter the preset safe process state under all candidate causal propagation relationships, and having available recovery resources that meet the most unfavorable recovery requirements.

[0117] After forming a recoverable state domain, based on the step-by-step application method of candidate execution increments, and utilizing the time delay range and response envelope of all candidate causal propagation relationships, the state range corresponding to each execution step and the delayed response stage after execution is completed is predicted, thus forming the execution state envelope.

[0118] When both the verification state envelope and the execution state envelope are located in the recoverable state domain, the corresponding candidate execution increment is retained; when either state envelope contains a state outside the recoverable state domain, the corresponding candidate execution increment is not retained.

[0119] For the retained candidate execution increments, the upper bounds of the same type of resource reservation requirements corresponding to each recovery start state within the verification state envelope and the execution state envelope are taken to form the overall resource reservation requirements, and the correspondence between the recovery start state range and the corresponding recovery increment sequence is recorded.

[0120] The recoverable state domain, recovery increment sequence, overall resource reservation requirements, and their corresponding relationships are associated and saved with the current causal network version used when the above content was formed. The recoverable state domain is not a component of the current causal network version.

[0121] Step S3 outputs the recoverable state domain, recovery increment sequence, overall resource reservation requirements, and candidate execution increments that pass the recoverability screening, all associated with the current causal network version.

[0122] S4. Combine the retained candidate execution increments with the corresponding paired verification sequences to form a two-stage adjustment unit. Determine the target adjustment unit based on the distinguishability of the candidate causal propagation relationship and the amount of reduction in prediction quality deviation.

[0123] For each two-stage adjustment unit, the expected verification results corresponding to each candidate causal propagation relationship are obtained, and a preset resolution scale is set for each component participating in the matching.

[0124] The preset resolution scales corresponding to the causal response components and the paired closure residuals are determined based on the sensor resolution, normal operating condition fluctuation scale, model prediction error, and allowable matching error, and have the same dimensions as the corresponding component components; the preset resolution scale corresponding to the selected alignment delay is represented by the number of sampling cycles; the preset resolution scale corresponding to the paired closure energy ratio is a dimensionless value.

[0125] When the expected verification results include vector sequences, the common valid sampling indices for each candidate causal propagation relationship are determined according to the same sensor channel order and relative sampling order, forming a common quantization index set. If the number of samples in the common quantization index set is lower than the preset minimum valid sampling number, the corresponding two-stage adjustment unit will not be used for target adjustment unit selection.

[0126] According to the corresponding preset resolution scale, the expected verification results of each candidate causal propagation relationship within the common quantization index set are quantized, and the quantization results are formed according to a fixed order of component parts. Candidate causal propagation relationships with the same quantization results are grouped into the same indistinguishable relationship group.

[0127] The number of candidate causal propagation relations contained in the indistinguishable relation group with the largest number of members is determined as the maximum size of the indistinguishable relation group for the corresponding two-stage adjustment unit. The smaller the maximum size of the indistinguishable relation group, the stronger the ability of the corresponding pairwise check sequence to distinguish candidate causal propagation relations.

[0128] Using the candidate causal propagation relationships in the current causal network version, predict the quality deviation after incremental step-by-step application of the candidate implementation. The difference between the quality deviation before application and the predicted quality deviation after application is determined as the reduction in predicted quality deviation under the corresponding candidate causal propagation relationship.

[0129] In this embodiment, the minimum value of the predicted quality deviation reduction under all candidate causal propagation relationships is taken as the predicted quality deviation reduction of the corresponding two-stage adjustment unit, so as to avoid overestimating the quality improvement effect when the actual causal propagation relationship has not yet been determined.

[0130] Select the two-stage adjustment unit with the smallest largest size of the indistinguishable relation group; if multiple two-stage adjustment units with the same largest size of indistinguishable relation groups exist, select the two-stage adjustment unit with the largest predicted quality deviation reduction as the target adjustment unit. If the predicted quality deviation reduction of all two-stage adjustment units is not greater than zero, the target adjustment unit is not determined for this round, and candidate execution increments are regenerated or the manufacturing process status is awaited for update.

[0131] Step S4 outputs the target adjustment unit, as well as the paired verification sequence associated with the target adjustment unit, the candidate execution increment, the expected verification result, the recovery increment sequence, and the overall resource reservation requirements.

[0132] S5. Apply the pairing verification sequence of the target adjustment unit and match the measured verification result with the expected verification result. When there is a unique match, apply the candidate execution increment step by step based on the corresponding candidate causal propagation relationship. When the recoverable state domain and recovery resource constraints are met during the step-by-step application, and the quality deviation is reduced within the corresponding time delay range after the application is completed, the candidate execution increment is determined as the formal process adjustment amount and the application is maintained. If any condition is not met, stop the execution and restore to the preset safe process state.

[0133] Before applying the target adjustment unit, establish the execution association between the target adjustment unit and the current causal network version, the recovery increment sequence, and the available recovery resources that meet the overall resource reservation requirements, and record the correspondence between each recovery start state range and the corresponding recovery increment sequence.

[0134] Based on the overall resource reservation requirements, the corresponding available recovery resources are locked. These locked available recovery resources are not allocated to other control tasks that may affect the recovery capability during this round of pairing verification sequence and candidate execution increment application.

[0135] The forward perturbation sequence and the reverse polarity inverse time perturbation sequence are applied sequentially according to the manufacturing cycle, and the measured response is collected according to the set of verification response channels, the order of sensing channels and the sampling period determined in step S2.

[0136] During the application of the pairing check sequence, the check state envelope is predicted in a rolling manner based on the current process state, the perturbations that have been applied and have not yet been applied, and the current causal network version. At each manufacturing cycle, it is determined whether the check state envelope is in the recoverable state domain and whether the available recovery resource margin meets the overall resource reservation requirements.

[0137] If any condition is not met, the remaining part of the pairing verification sequence is terminated, and the corresponding recovery increment sequence is called according to the recovery start state range in which the current state is located, so that the manufacturing process enters the preset safe process state.

[0138] After the pairing verification sequence is applied, the measured verification results corresponding to each candidate causal propagation relationship are formed according to the predicted baseline response subtraction, reverse time-series rearrangement of the antipolar response, time delay range discretization, pairing closure energy ratio calculation and pairing decomposition rules in step S2.

[0139] The use of the same time delay alignment method means using the same time delay discretization, effective sampling index determination, paired closure energy ratio calculation and minimum value selection rules as in step S2, and does not mean directly using the selected alignment time delay in the expected verification result.

[0140] Under the selected alignment delay in the actual measurement, half of the difference between the measured positive response increment vector and the measured reverse polarity response increment vector after alignment is used to form the measured causal response component, and half of the sum of the two is used to form the measured paired closure residual. The measured selected alignment delay, the measured paired closure energy ratio, the measured causal response component, and the measured paired closure residual are used to form the measured verification result.

[0141] According to the common quantization index set, preset resolution scale and quantization rules adopted in step S4, the measured verification results corresponding to each candidate causal propagation relationship are quantitatively matched with the expected verification results.

[0142] When the measured verification result corresponding to the candidate causal propagation relationship is the same as the expected verification result, and the difference of each component does not exceed the corresponding preset resolution scale, the candidate causal propagation relationship is determined to meet the matching condition.

[0143] If only one candidate causal propagation relationship satisfies the matching condition, it is determined as the basis for execution; if no unique match can be found, no candidate execution increment is applied, and the corresponding recovery increment sequence is called to bring the manufacturing process into the preset safe process state.

[0144] like Figure 4 As shown, after determining the execution basis, the system applies candidate execution increments step by step according to the manufacturing cycle, and continuously monitors the execution status envelope and recovery resource constraints. When the quality improvement conditions are met, the process adjustment amount is maintained, and when the conditions are not met, a safe recovery is performed.

[0145] Once the execution basis is definitively determined, verify that the current causal network version, target adjustment unit, recovery increment sequence, and locked available recovery resources are all consistent with the execution association. If the verification passes, apply candidate execution increments step by step according to the manufacturing cycle.

[0146] During the step-by-step application, based on the candidate causal propagation relationship corresponding to the execution basis, the execution state envelope is predicted in a rolling manner according to the current process state, the execution increments that have been applied and those that have not yet been applied, and at each execution step, it is determined whether the execution state envelope is in the recoverable state domain and whether the available recovery resource margin meets the overall resource reservation requirements.

[0147] If any condition is not met, stop applying the remaining part of the candidate execution increment and call the corresponding recovery increment sequence according to the current state.

[0148] After the candidate execution increment is applied in steps, the quality confirmation time interval is determined according to the time delay range corresponding to the execution basis, and the same quality index order, scale conversion method and quality deviation aggregation rule as in step S1 are used to determine the quality deviation value after application.

[0149] If the applied quality deviation value is lower than the pre-application quality deviation value within the corresponding time delay range, the candidate execution increment is determined as the formal process adjustment amount, and the adjusted controllable process parameter value is used as the current process parameter value or new adjustment benchmark for the subsequent manufacturing process; if no decrease in quality deviation value is detected, the corresponding recovery increment sequence is called to make the manufacturing process enter the preset safe process state.

[0150] After comparing the measured verification results with the expected verification results, the abnormal observations are determined based on the deviations between the measured causal response components and the measured paired closure residuals and the corresponding expected components.

[0151] When the measured verification result can be uniquely matched, the uniquely matched candidate causal propagation relationship is used as the reference relationship for anomaly judgment. When a unique match cannot be made, the differences between the measured verification result and each expected verification result are normalized according to the corresponding preset resolution scale, and the candidate causal propagation relationship with the smallest overall deviation is selected as the anomaly diagnosis reference relationship; the anomaly diagnosis reference relationship is only used for anomaly classification and is not used as the basis for executing candidate execution increments.

[0152] Using the expected component corresponding to the reference relationship as the comparison object, when the deviation of the measured causal response component exceeds the corresponding preset resolution scale, and the deviation of the measured paired closure residual does not exceed the corresponding preset resolution scale, the abnormal observation item is identified as a candidate causal propagation relationship change.

[0153] When the deviation of the measured paired closure residual exceeds the corresponding preset resolution scale, the abnormal observation is identified as an execution measurement chain anomaly. The expected paired closure residual already includes predictable nonlinearities, hysteresis, and normal operating condition fluctuations within the current response envelope; therefore, the deviation represents an additional non-closed response beyond the expected closure range. When the deviations of both the causal response component and the paired closure residual exceed the corresponding preset resolution scale, it is treated as an execution measurement chain anomaly.

[0154] In response to changes in candidate causal propagation relationships, the delay range or response envelope of the corresponding candidate causal propagation relationship is corrected based on the measured alignment delay and measured causal response components. The secure and maintainable state domain is re-formed according to step S1, and the recoverable state domain, recovery increment sequence, and resource reservation requirements associated with the corrected current causal network version are re-formed according to step S3.

[0155] When both the re-formed safe and maintainable state domain and the recoverable state domain are not empty, an updated version of the current causal network is formed, and the recoverable state domain associated with this current causal network version is updated; when either state domain is empty, or when an abnormal observation is determined to be an execution measurement chain anomaly, the controllable process parameters are switched to the preset safe process parameter group.

[0156] After this round of execution is completed, unlock the available recovery resources that are no longer needed, and record the target adjustment unit, matching results, candidate execution increment execution status, quality confirmation results, abnormal observations and the current causal network version for use in the next round of manufacturing process optimization decision-making.

[0157] like Figure 2 As shown, another embodiment of the present invention also provides a manufacturing process optimization decision system oriented towards quality objectives, for executing the manufacturing process optimization decision method oriented towards quality objectives as described above, including:

[0158] The data modeling module is used to acquire controllable process parameters, sensor time series data, quality targets, safety boundaries and available recovery resources, determine quality deviations, construct a time series causal network, set multiple candidate causal propagation relationships and their time delay range and response envelope, and form the current causal network version and the safe and maintainable state domain.

[0159] The verification generation module is used to generate candidate execution increments. It generates a paired verification sequence consisting of positive perturbations and anti-polarity inverse time-series perturbations with an increment algebraic sum of zero in the controllable process parameter channel. It predicts the paired response of each candidate causal propagation relationship and forms the expected verification result containing causal response components and paired closure residuals after time delay alignment and paired decomposition.

[0160] The recovery filtering module is used to form a recoverable state domain based on the recovery requirements and available recovery resources when the pairing verification is aborted and the candidate execution increment fails. It retains candidate execution increments whose verification state envelope and execution state envelope are located in the recoverable state domain, whose recovery state envelope is located in the safe and maintainable state domain, and whose manufacturing process can be restored to the preset safe process state.

[0161] The adjustment unit determination module is used to form a two-stage adjustment unit by combining the retained candidate execution increments with the corresponding paired verification sequences, and to determine the target adjustment unit based on the distinguishability of the candidate causal propagation relationship and the amount of reduction in prediction quality deviation.

[0162] The execution control module is used to apply the pairing verification sequence of the target adjustment unit and match the measured verification result with the expected verification result. When there is a unique match, the candidate execution increment is applied step by step based on the corresponding candidate causal propagation relationship. When the recoverable state domain and recovery resource constraints are met during the step-by-step application, and the quality deviation is reduced within the corresponding time delay range after the application is completed, the candidate execution increment is determined as the formal process adjustment amount and the application is maintained. If any condition is not met, the execution is stopped and the manufacturing process is restored to the preset safe process state.

[0163] In this embodiment, the data modeling module, verification generation module, recovery screening module, adjustment unit determination module, and execution control module are respectively used to execute the aforementioned steps S1, S2, S3, S4, and S5. Regarding the data acquired by each module, the processing procedures performed, the judgment rules adopted, and the processing results formed, please refer to the corresponding content in the aforementioned method embodiments, which will not be repeated here. The above modules are logical modules divided according to function, and can be implemented by independent processing units or by the processor calling program instructions stored in memory. The modules exchange the current causal network version, candidate execution increments, paired verification sequences, expected verification results, recoverable state domains, recovery increment sequences, resource reservation requirements, and target adjustment units to execute the manufacturing process optimization decision-making method oriented towards quality objectives.

[0164] In summary, this invention addresses the problem of dynamic changes in the relationship between process parameters, process states, and quality objectives in complex manufacturing processes. It proposes a manufacturing process optimization decision-making method and system oriented towards quality objectives, forming a complete technical solution from process state perception, causal relationship analysis, optimization decision generation to execution feedback control. By comprehensively processing multi-source information in the manufacturing process, it achieves synergy between quality objective-driven process optimization decision-making and process safety control.

[0165] This invention can be applied to manufacturing process scenarios with multiple parameter coupling, dynamic changes, and high quality control requirements. It provides technical support for process optimization, quality improvement, and stable operation control in intelligent manufacturing, and has good engineering application value and promotion prospects.

[0166] In the description of this specification, the use of terms such as "one embodiment," "some embodiments," "example," and "specific example" indicates that a specific feature, structure, or technical solution described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the described specific features, structures, or technical solutions can be combined in any suitable manner in one or more embodiments or examples. Without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the technical features of those different embodiments or examples.

[0167] The steps shown in the flowchart in this application can be implemented by computer program instructions, control logic or corresponding functional modules. The execution order of each step can be adjusted according to actual application requirements. Other equivalent methods can be used without affecting the implementation effect of the technical solution.

[0168] The above description is merely a specific embodiment of this application, and the scope of protection of this application is not limited to the above-described embodiments. Any equivalent substitutions or modifications made by those skilled in the art based on the disclosure of this application without departing from the technical concept of 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 defined in the claims.

Claims

1. A manufacturing process optimization decision-making method oriented towards quality objectives, characterized in that, Includes the following steps: Acquire controllable process parameters, sensor timing data, quality targets, safety boundaries, and available recovery resources to determine quality deviations; Construct a temporal causal network, set multiple candidate causal propagation relationships, time delay ranges, and response envelopes, and form the current causal network version and a secure and maintainable state domain; Candidate execution increments are generated. In the controllable process parameter channel, a paired verification sequence consisting of positive perturbation and anti-polarity inverse time-series perturbation with an increment algebraic sum of zero is generated. The paired response of each candidate causal propagation relationship is predicted. After time delay alignment and paired decomposition, the expected verification result containing causal response components and paired closure residuals is formed. A recoverable state domain is formed based on the recovery requirements and available recovery resources when the pairing verification is aborted and the candidate execution increment fails. Candidate execution increments whose verification state envelope and execution state envelope are located in the recoverable state domain, whose recovery state envelope is located in the safe and maintainable state domain, and whose manufacturing process can be restored to the preset safe process state are retained. The retained candidate execution increments and corresponding paired verification sequences are combined to form a two-stage adjustment unit. The target adjustment unit is determined based on the distinguishability of the candidate causal propagation relationship and the amount of reduction in prediction quality deviation. Apply the pairing verification sequence of the target adjustment unit to match the measured verification result with the expected verification result; when there is a unique match, apply the candidate execution increment step by step based on the corresponding candidate causal propagation relationship; when the recoverable state domain and recovery resource constraints are met during the step-by-step application, and the quality deviation is reduced within the corresponding time delay range after the application is completed, the candidate execution increment is determined as the formal process adjustment amount and the application is maintained; when any condition is not met, the execution is stopped and the system is restored to the preset safe process state.

2. The manufacturing process optimization decision-making method oriented towards quality objectives according to claim 1, characterized in that, The current causal network version and the secure, maintainable state domain are formed as follows: Controllable process parameters, process states represented by sensing time-series data, and quality indicators corresponding to quality targets are set as network nodes. Directed propagation edges are established according to the direction of influence transmission. Multiple candidate causal propagation relationships are generated for each directed propagation edge according to process constraints. The time delay range and response envelope of each candidate causal propagation relationship are set. The set of states that satisfy the safety boundaries and equipment constraints is used as the initial set of states; For each state to be screened, the next state envelope corresponding to the increment of each controllable process parameter that satisfies the equipment constraints is predicted by using the time delay range and response envelope of each candidate causal propagation relationship. When there exists a controllable process parameter increment that makes the next state envelope corresponding to all candidate causal propagation relationships within the current round's state set, the state to be screened is retained. If no more states are eliminated in the current round of iterative filtering, the set of retained states is determined as the safe and maintainable state domain. Associating the temporal causal network, candidate causal propagation relationships, corresponding time delay ranges and response envelopes, and secure and maintainable state domains with the same version identifier forms the current causal network version.

3. The manufacturing process optimization decision-making method oriented towards quality objectives according to claim 1, characterized in that, The process of generating the pairwise verification sequence and predicting the paired responses for each candidate causal propagation relationship includes: A forward perturbation sequence is generated in the same controllable process parameter channel. The perturbation quantities in the forward perturbation sequence are inverted in the opposite time sequence to form an anti-polarity inverse time sequence perturbation sequence, so that the algebraic sum of the controllable process parameter increments corresponding to the forward perturbation sequence and the anti-polarity inverse time sequence perturbation sequence is zero. For each candidate causal propagation relationship in the current causal network version, based on the time delay range and response envelope, the sensor responses caused by the forward perturbation sequence and the reverse polarity reverse time-series perturbation sequence are predicted respectively. The predicted baseline response of the same manufacturing cycle when no corresponding perturbation sequence is applied is subtracted respectively. The predicted responses corresponding to the reverse polarity reverse time-series perturbation are rearranged in reverse time order without changing the response polarity, and the predicted forward response increment sequence and the predicted reverse polarity response increment sequence are formed according to the same sensor channel order. The forward perturbation sequence, the reverse polarity inverse time-series perturbation sequence, and the state envelopes corresponding to the connection process of the two sequences are used to form the verification state envelope. Paired verification sequences whose verification state envelopes are located in the safe and maintainable state domain and whose quality perturbations do not exceed the verification tolerance are retained. The retained paired verification sequences are associated with the corresponding candidate execution increments.

4. The manufacturing process optimization decision-making method oriented towards quality objectives according to claim 3, characterized in that, The expected verification result is generated through the following steps: The time delay range of each candidate causal propagation relationship is discretized according to the sampling period of the sensing time series data; For each discrete time delay, determine the effective sampling index set with corresponding sample values ​​for both the predicted positive response increment sequence and the predicted reverse polarity response increment sequence. Form an effective candidate aligned time delay set by the discrete time delays with the effective sampling number reaching the preset minimum effective sampling number, and retain the pairing verification sequence where the effective candidate aligned time delay set of each candidate causal propagation relationship is not empty. Construct an invertible diagonal response scale matrix based on the positive-zero response scales of each sensor channel, following the same sensor channel sequence. Candidate causal propagation relationships are calculated using the following formula. Candidate alignment delay The following are the pairing closure energy ratios: ； In the formula, As a candidate causal propagation relationship identifier, The candidate alignment delay is expressed in terms of the number of sampling beats. Candidate alignment delay The corresponding set of valid sampling indices, To determine the effective number of samples, For effective sampling index, For the first The predicted positive response increment vector for each sampling tick. For reverse time rearrangement and candidate alignment delay Aligned predicted reverse polarity response increment vectors The response scaling matrix has diagonal elements that have the same dimensions as the corresponding sensing response. Assuming a dimensionless positive number, The ratio of paired closure energies. Represents the 2-norm of a vector; The selected alignment delay is determined from the candidate alignment delays with the smallest paired closure energy ratio; With a selected alignment delay, half of the difference between the predicted positive response increment vector corresponding to each effective sampling index and the aligned predicted reverse polarity response increment vector is used to form the causal response component. Half of the sum of the two is used to form the paired closure residual. The selected alignment delay, the corresponding paired closure energy ratio, the causal response component, and the paired closure residual are used to form the expected verification result of the corresponding candidate causal propagation relationship.

5. The manufacturing process optimization decision-making method oriented towards quality objectives according to claim 1, characterized in that, Forming the recoverable state domain and retaining the candidate execution increments includes: The states that may be reached when the pairing verification sequence is terminated and the states that may be reached when the candidate execution incremental step-by-step application fails are determined as the recovery start state; For each recovery start state, determine the same recovery increment sequence that satisfies the device constraints. Based on the time delay range and response envelope of all candidate causal propagation relationships, predict the recovery state envelope and recovery requirements of various recovery resources corresponding to the same recovery increment sequence. Determine resource reservation requirements based on the upper bound of the recovery requirements of various recovery resources under each candidate causal propagation relationship. Under all candidate causal propagation relationships, the recovery state envelope remains in the safe and maintainable state domain. The recovery increment sequence enables the manufacturing process to enter a preset safe process state. When the available recovery resources meet the resource reservation requirements, the corresponding recovery start state is included in the recoverable state domain. Candidate execution increments whose verification state envelope and execution state envelope are both located in the recoverable state domain are retained.

6. The manufacturing process optimization decision-making method oriented towards quality objectives according to claim 5, characterized in that, The target adjustment unit is determined according to the following rules: The retained candidate execution increments are combined with the corresponding paired verification sequences to form a two-stage adjustment unit set; According to the preset resolution scale corresponding to each component of the expected verification result, the expected verification result corresponding to each candidate causal propagation relationship under each two-stage adjustment unit is quantified, and the candidate causal propagation relationships with the same quantification result are divided into the same indistinguishable relationship group. The number of candidate causal propagation relations contained in the indistinguishable relation group with the largest number of members is determined as the maximum size of the indistinguishable relation group corresponding to the two-stage adjustment unit, and the amount of reduction in prediction quality deviation is determined based on the prediction quality deviation before and after the application of the candidate execution increment. Select the two-stage adjustment unit with the largest and smallest size of the indistinguishable relation group; When there are multiple two-stage adjustment units with the same maximum size for indistinguishable relational groups, the two-stage adjustment unit with the largest reduction in prediction quality deviation is selected as the target adjustment unit.

7. The manufacturing process optimization decision-making method oriented towards quality objectives according to claim 6, characterized in that, The process of applying the pairing check sequence to the target adjustment unit includes: Establish execution association relationships between the target adjustment unit and the current causal network version, the corresponding recovery increment sequence, and the available recovery resources that meet the resource reservation requirements, and lock the available recovery resources corresponding to the execution association relationships during the pairing verification sequence and the application of candidate execution increments; According to the manufacturing rhythm, the forward perturbation sequence and the reverse polarity inverse time perturbation sequence are applied sequentially. The measured response of each sensing channel is collected. During the application of the pairing verification sequence, the verification state envelope is predicted based on the candidate causal propagation relationship in the current causal network version. When the verification state envelope exceeds the recoverable state domain or the available recovery resource margin does not meet the resource reservation requirements, the pairing verification sequence is stopped, and the recovery increment sequence corresponding to the execution association relationship is applied. After the paired verification sequence is applied, the predicted baseline response of the same manufacturing cycle when no corresponding perturbation sequence is applied is subtracted. The measured responses corresponding to the reverse polarity reverse time-series perturbation are rearranged in reverse time sequence without changing the response polarity to form the measured forward response increment sequence and the measured reverse-order reverse polarity response increment sequence. According to the time delay alignment and paired decomposition method used when forming the expected verification results, the measured causal response components and measured paired closure residuals corresponding to each candidate causal propagation relationship are formed, and the measured verification results are composed.

8. The manufacturing process optimization decision-making method oriented towards quality objectives according to claim 7, characterized in that, Based on the measured verification results, the execution criteria are determined and the candidate execution increments are applied step by step: According to the preset resolution scale, the measured verification results are quantitatively matched with the expected verification results corresponding to each candidate causal propagation relationship; when only one candidate causal propagation relationship satisfies the same quantification result and the difference between each component does not exceed the corresponding preset resolution scale, the corresponding candidate causal propagation relationship is determined as the execution basis; when a unique match cannot be made, no candidate execution increment is applied, and a recovery increment sequence corresponding to the execution association relationship is applied. When the current causal network version, target adjustment unit, recovery increment sequence, and locked available recovery resources associated with the execution relationship remain unchanged, candidate execution increments are applied step by step according to the manufacturing cycle, and the execution state envelope is predicted based on the execution basis and the current process state; during the step-by-step application, if the execution state envelope exceeds the recoverable state domain or the available recovery resource margin does not meet the resource reservation requirements, the application of the remaining part of the candidate execution increments is stopped, and the recovery increment sequence is applied. After the candidate execution increment is applied in steps, if a decrease in quality deviation is detected within the time delay range corresponding to the execution basis compared to before the application, the candidate execution increment is determined as the formal process adjustment amount and is maintained; if no decrease in quality deviation is detected, the recovery increment sequence is applied.

9. The manufacturing process optimization decision-making method oriented towards quality objectives according to claim 8, characterized in that, The current causal network version is updated based on anomaly observations in the following manner: Anomalies are determined based on the deviations between measured causal response components and measured paired closure residuals and corresponding expected components. When the deviation of the causal response component exceeds the corresponding preset resolution scale and the deviation of the paired closure residual does not exceed the corresponding preset resolution scale, the abnormal observation is identified as a candidate causal propagation relationship change. When the deviation of the paired closure residual exceeds the corresponding preset resolution scale, the abnormal observation item is identified as an abnormality in the execution measurement chain; In response to changes in candidate causal propagation relationships, the time delay range or response envelope of the corresponding candidate causal propagation relationships is modified, and a safe and maintainable state domain and a recoverable state domain are re-formed. The current causal network version is updated when both the safe and maintainable state domain and the recoverable state domain are not empty. When an abnormal observation is identified as an execution measurement chain anomaly, or when the newly formed safe and maintainable state domain or recoverable state domain is empty, the controllable process parameters will be switched to the preset safe process parameter group.

10. A manufacturing process optimization decision system oriented towards quality objectives, used to execute the manufacturing process optimization decision method oriented towards quality objectives as described in any one of claims 1-9, characterized in that, include: The data modeling module is used to acquire controllable process parameters, sensor time series data, quality targets, safety boundaries and available recovery resources, determine quality deviations, construct a time series causal network, set multiple candidate causal propagation relationships and their time delay range and response envelope, and form the current causal network version and the safe and maintainable state domain. The verification generation module is used to generate candidate execution increments. It generates a paired verification sequence consisting of positive perturbations and anti-polarity inverse time-series perturbations with an increment algebraic sum of zero in the controllable process parameter channel. It predicts the paired response of each candidate causal propagation relationship and forms the expected verification result containing causal response components and paired closure residuals after time delay alignment and paired decomposition. The recovery filtering module is used to form a recoverable state domain based on the recovery requirements and available recovery resources when the pairing verification is aborted and the candidate execution increment fails. It retains candidate execution increments whose verification state envelope and execution state envelope are located in the recoverable state domain, whose recovery state envelope is located in the safe and maintainable state domain, and whose manufacturing process can be restored to the preset safe process state. The adjustment unit determination module is used to form a two-stage adjustment unit by combining the retained candidate execution increments with the corresponding paired verification sequences, and to determine the target adjustment unit based on the distinguishability of the candidate causal propagation relationship and the amount of reduction in prediction quality deviation. The execution control module is used to apply the pairing verification sequence of the target adjustment unit and match the measured verification result with the expected verification result. When there is a unique match, the candidate execution increment is applied step by step based on the corresponding candidate causal propagation relationship. When the recoverable state domain and recovery resource constraints are met during the step-by-step application, and the quality deviation is reduced within the corresponding time delay range after the application is completed, the candidate execution increment is determined as the formal process adjustment amount and the application is maintained. If any condition is not met, the execution is stopped and the manufacturing process is restored to the preset safe process state.

Images