Self-adaptive adjustment method and system for unstacking height of stamping robot

Abstract

The invention discloses a self-adaptive adjustment method and system for the unstacking height of a stamping robot, and relates to the technical field of intelligent manufacturing, and the method comprises the steps: obtaining multi-modal data, extracting a root cause node of the multi-modal data through a recursive backtracking causal chain, and generating an anti-interference rule library based on the interference condition characteristics of the root cause node and a compensation strategy mapping relation; generating a decision vector through a causal entropy balance target of a neural symbol inference engine; mapping the decision vector into a discrete event model, and verifying the time sequence security through a linear temporal logic protocol; monitoring an actual causal entropy value generated by the decision vector, comparing the actual causal entropy value with a causal entropy balance target, and when the actual causal entropy value deviates from a tolerance value of the causal entropy balance target, marking the actual causal entropy value as a high-risk signal; and triggering root cause backtracking analysis based on the high-risk signal, and if a backtracking result is confirmed as a rule defect, updating a related rule set in the rule base.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and in particular to a method and system for adaptive adjustment of the destacking height of a stamping robot. Background Technology

[0002] In modern stamping production lines, industrial robots are widely used in sheet metal destacking operations. Their core task is to stably, efficiently, and accurately grasp individual sheets from stacked stacks and transport them to subsequent stamping presses or alignment tables. Precise positioning of the destacking height is one of the key aspects ensuring the smooth operation of this process. However, in actual production environments, the destacking height is not fixed but is affected by various dynamic factors, posing a significant challenge to traditional pre-set fixed-height destacking methods. Existing technologies suffer from a lack of dynamic interference tracing, static and rigid rule bases, and an imbalance between neural decision-making and symbolic security, resulting in a destacking success rate of only 82%. Therefore, there is an urgent need for an intelligent adjustment method with real-time multi-source interference tracing, dynamic rule base evolution, and neural-symbolic collaborative decision-making capabilities to achieve adaptive and precise control of the destacking height under complex working conditions, thereby overcoming the bottlenecks of existing technologies. Summary of the Invention

[0003] This application provides a method and system for adaptive adjustment of the destacking height of a stamping robot, which solves the temporal security problem caused by the lack of dynamic interference tracing leading to rigid rule base and the separation of neural decision-making and symbolic verification in the prior art, and achieves highly adaptive and precise control in a multi-source coupled interference field.

[0004] This application provides a method for adaptive adjustment of the depalletizing height of a stamping robot, including:

[0005] S1: Acquire multimodal data, extract the root cause nodes of the multimodal data by recursively backtracking the causal chain, and generate an anti-interference rule base based on the interference condition characteristics and compensation strategy mapping relationship of the root cause nodes.

[0006] S2: Based on the anti-interference rule base and multimodal data, a decision vector is generated through the causal entropy balance target of the neural symbolic reasoning engine; the decision vector is mapped to the discrete event model, and the temporal safety is verified by linear temporal logic reduction. If the verification fails, the safety policy is activated to correct the decision.

[0007] S3: Monitor the actual causal entropy value generated by the decision vector and compare it with the causal entropy balance target. When the actual causal entropy value deviates from the tolerance value of the causal entropy balance target, it is marked as a high-risk signal.

[0008] S4: Based on the backtracking analysis of the root cause triggered by the high-risk signal, if the backtracking result confirms that it is a rule defect, the relevant rule set in the rule base is updated; if the backtracking result is not a rule defect, the compensation strategy is activated according to the fault type.

[0009] Furthermore, the multimodal data includes: equipment body parameters, process characteristic parameters, spatiotemporal interference parameters, physical coupling parameters, cooperative state parameters, and expert rule data;

[0010] The equipment body parameters are the real-time status parameters of joint torque and motor current recorded in the equipment operation log.

[0011] The process characteristic parameters are the process parameters of die size and stamping speed in the stamping process database;

[0012] The spatiotemporal interference parameters are the workspace overlap and clock synchronization deviation rate collected by the sensor network.

[0013] The physical coupling parameters are process parameter coupling data that include the vibration transmission coefficient and the gas source fluctuation coefficient;

[0014] The collaborative status parameters are the local fault codes, neighboring machine operating status, and overall production capacity margin collected by the collaborative sensing nodes.

[0015] The expert rule data refers to historical fault handling rules provided by expert experience documents.

[0016] Furthermore, the recursive backtracking causal chain includes: starting from the current abnormal event, constructing a fault propagation topology graph based on multimodal data; fusing spatiotemporal interference parameters and physical coupling parameters through the fault propagation topology graph to generate a causal field; performing reverse traversal along the causal propagation topology manifold in the causal field; and deleting unnecessary intermediate variables based on the confidence level of historical handling cases; performing equipment physical boundary condition verification on candidate nodes; performing elimination operations on nodes that exceed the physical boundary; stopping backtracking when the node's influence is lower than the threshold defined in the expert experience document or reaches the initial boundary node of the process system; and outputting a set of root cause nodes.

[0017] Furthermore, the anti-interference rule base includes: an interference compensation rule set, physical constraint clauses, and historical confidence level labels;

[0018] The interference compensation rule set is a dynamic mapping relationship between the interference condition characteristics and the compensation rules established based on the root cause node.

[0019] The physical constraint clauses are the device physical boundary constraints built into each rule;

[0020] The historical confidence label is the confidence value associated with the historical success rate of each rule.

[0021] Furthermore, the decision vector includes: destacking height adjustment amount, vibration anti-phase control spectrum, and safety margin indicator;

[0022] The stacking height adjustment amount is a Z-axis travel compensation value generated by the neural symbolic reasoning engine based on the anti-interference rule base;

[0023] The vibration anti-phase control spectrum is a set of frequency domain waveform parameters used to actively cancel vibration interference;

[0024] The safety margin indicator is a scalar value that identifies the safety boundary of a decision.

[0025] Furthermore, the safety strategy includes: reordering the decision action sequence that violates the linear temporal logic specification, and inserting safety checkpoints in the gaps between the reordered actions; when a safety checkpoint is triggered, compressing the stacking height adjustment amount in the decision vector to within the allowable safety margin range, and modifying the vibration antiphase control spectrum parameters to avoid the resonance frequency band.

[0026] Furthermore, the root cause backtracking analysis includes: reconstructing the causal field by fusing the latest spatiotemporal interference and physical coupling parameters to form a dynamically evolving causal transmission network; traversing backward along the reconstructed causal transmission network, filtering effective paths by combining historical confidence labels in the rule base, and performing secondary verification of the physical boundary conditions of the retained nodes; if new cause nodes not covered by the rule base or low-confidence rules are found, they are determined to be rule defects; if all nodes pass the verification and conform to the rule base logic, they are determined to be non-rule defects and a specific fault type classification is output.

[0027] Furthermore, the causal field also includes a collaborative causal field for multi-device collaborative control: a collaborative causal field containing a causal transmission topology manifold is constructed based on spatiotemporal interference parameters and physical coupling parameters; by recursively backtracking the chain of abnormal events of multiple devices, a minimum complete set of interference rules is extracted from the collaborative causal field and continuously updated, and this set of rules is embedded as a sub-module into the anti-interference rule base; when the collaborative causal field detects that the anomaly occurrence rate is greater than the anomaly threshold, the causal transmission topology is reconstructed in real time, the reconstructed path is injected into the neural symbolic inference engine, collaborative control instructions are generated by fusion weight balancing, and used as a component of the decision vector.

[0028] Furthermore, the collaborative causal field also includes: constructing a collaborative fault-tolerant control field based on multimodal data; synchronously executing the acquisition of vibration spectrum of the faulty machine to generate an anti-phase wave for cancellation by the neighboring machine in the collaborative fault-tolerant control field; simultaneously monitoring the reverse offset triggered when the overlap of the working area exceeds 40%; classifying the collaborative handling of vibration interference and spatial conflict; verifying whether the results simultaneously meet the dual requirements of a production capacity recovery rate greater than 95% and an interference elimination degree greater than 90%; and updating the fault-tolerant rules of the qualified cases to the anti-interference rule base by generating fault-tolerant rules through a reverse growth aggregation algorithm.

[0029] An adaptive adjustment system for destacking height of a stamping robot, the system comprising:

[0030] Multimodal data acquisition module: Real-time acquisition of equipment body parameters, process characteristic parameters, spatiotemporal interference parameters, physical coupling parameters, collaborative state parameters and expert rule data;

[0031] Recursive backtracking analysis module: Constructs a fault propagation topology graph based on multimodal data, integrates spatiotemporal interference and physical coupling parameters to generate a causal field, and traverses the causal propagation topology manifold in reverse to extract the set of root cause nodes;

[0032] Rule base construction module: Generates an anti-interference rule base based on the root cause node, which includes interference compensation rule set, physical constraint clauses and historical confidence labels;

[0033] The neural symbolic decision-making module generates decision vectors through the neural symbolic reasoning engine, performs linear temporal logic reduction verification, and activates security policy correction decisions when verification fails.

[0034] Causal Entropy Monitoring Module: Monitors the actual causal entropy value generated by the decision vector and marks a high-risk signal when it deviates from the target tolerance value;

[0035] Cooperative causal field processing module: Constructs a cooperative causal field based on spatiotemporal interference and physical coupling parameters, extracts the minimum complete interference rule set and embeds it into the rule base, and reconstructs the topology path to generate cooperative control instructions when anomalies exceed the limit;

[0036] Fault-tolerant control execution module: Synchronously executes vibration anti-phase cancellation and overlap-triggered reverse offset in the collaborative fault-tolerant control field, and handles collaborative issues in a hierarchical manner through the resource rescheduling engine;

[0037] Closed-loop verification module: Verifies the achievement of dual targets of capacity recovery rate ≥95% and interference elimination degree ≥90%;

[0038] Rule self-evolution module: Transforms compliant cases into fault-tolerant rules through a reverse growth aggregation algorithm and updates the anti-interference rule base in real time.

[0039] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0040] By adopting a collaborative architecture of recursive backtracking causal chain and neural symbolic reasoning engine, the root cause tracing mechanism significantly improves the accuracy of interference localization and effectively eliminates redundant intermediate variables; the neural symbolic reasoning engine greatly enhances the temporal security verification capability through the advantages of dynamic balancing rule constraints and data fitting; the collaborative causal field reconstruction technology realizes the synchronous resolution of multi-machine vibration interference and spatial conflict, and drives the rule base to continuously self-evolve, achieving significant optimization of the accuracy of adaptive adjustment of stacking height. Attached Figure Description

[0041] Figure 1 This is a flowchart of the adaptive adjustment method for the destacking height of the stamping robot in an embodiment of the present invention;

[0042] Figure 2 This is a system architecture diagram of the stamping robot's stacking height adaptive adjustment system in an embodiment of the present invention. Detailed Implementation

[0043] To facilitate understanding of the present invention, a more complete description of this application will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the present invention.

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention; the term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0045] Example 1: As Figure 1 As shown, there is an adaptive adjustment method for the destacking height of the stamping robot.

[0046] S1: Acquire multimodal data, extract the root cause nodes of the multimodal data by recursively backtracking the causal chain, and generate an anti-interference rule base based on the interference condition characteristics and compensation strategy mapping relationship of the root cause nodes.

[0047] The multimodal data includes: equipment body parameters, process characteristic parameters, spatiotemporal interference parameters, physical coupling parameters, cooperative state parameters, and expert rule data;

[0048] The equipment body parameters are the real-time status parameters of joint torque and motor current recorded in the equipment operation log.

[0049] The process characteristic parameters are the process parameters of die size and stamping speed in the stamping process database;

[0050] The spatiotemporal interference parameters are the workspace overlap and clock synchronization deviation rate collected by the sensor network.

[0051] The physical coupling parameters are process parameter coupling data that include the vibration transmission coefficient and the gas source fluctuation coefficient;

[0052] The collaborative status parameters are the local fault codes, neighboring machine operating status, and overall production capacity margin collected by the collaborative sensing nodes.

[0053] The expert rule data refers to historical fault handling rules provided by expert experience documents.

[0054] Specifically, real-time data such as joint torque and motor current collected by the equipment controller are used to characterize the real-time operating load characteristics of the machine body; process benchmark parameters such as mold size and stamping speed extracted from the stamping process database are used to define the constraint boundaries of the working environment; workspace overlap and cycle synchronization deviation rate based on the fusion perception of lidar and encoder are used to identify the risk of multi-device collaborative conflicts; vibration transmission coefficient collected by vibration sensor network and air source fluctuation coefficient collected by air pressure sensor are used to model the energy transfer path between equipment; local fault codes, neighboring machine working status, and overall line capacity margin are uploaded in real time by distributed collaborative sensing nodes; and a structured storage of historical fault handling rules is used, with each rule associated with confidence weight and applicable working condition label.

[0055] The recursive backtracking causal chain includes: starting from the current abnormal event, constructing a fault propagation topology graph based on multimodal data; fusing spatiotemporal interference parameters and physical coupling parameters through the fault propagation topology graph to generate a causal field; performing a reverse traversal along the causal transmission topology manifold in the causal field; and deleting unnecessary intermediate variables based on the confidence level of historical handling cases; performing equipment physical boundary condition verification on candidate nodes; performing a removal operation on nodes that exceed the physical boundary; stopping the backtracking when the node's influence is below the threshold defined in the expert experience document or reaches the initial boundary node of the process system; and outputting a set of root cause nodes.

[0056] Specifically, starting with the current abnormal event (such as Z-axis positioning deviation > 5mm), a fault propagation topology is constructed by fusing multimodal data: ,in, This is a fault propagation topology diagram. It is a vertex set, containing device nodes, process nodes, and environment nodes. For the edge set, the propagation path of the fault among multiple nodes is represented. A tensor fusion algorithm maps the spatiotemporal interference parameters and physical coupling parameters to a unified field space.

[0057] ,

[0058] in, For causal field tensor, The sigmoid activation function is used to compress the output into a probability space. This is a tensor splicing operation. Let be the spacetime interference tensor, with a value range of . (Spatiotemporal interference parameter matrix, where m and n are the number of rows and columns of the matrix); For the physical coupling tensor, the value range is... (Physical coupling parameter matrix, where p and q are the number of rows and columns of the matrix); The weight matrix of the spatiotemporal interferometry parameter matrix has a value range of 1. k is the number of hidden layer nodes in the neural network, here k=256, taking 256-dimensional features; The weight matrix is ​​the physical coupling parameter matrix, and its values ​​range from 1 to 2. The final output contains the causal transmission topological manifold M.

[0059] Traverse backwards along M from the abnormal event node to the initial boundary node, deleting intermediate variables with a confidence level of <0.7 in historical handling cases; verify the physical boundary conditions for candidate nodes, ensuring that the joint torque is less than or equal to the design limit value and the equipment frequency is within ±5Hz of the equipment's natural frequency; if a node exceeds the limits, it is immediately removed.

[0060] Calculate the current candidate node Contribution to overall failure and compared with the threshold preset in the expert experience document. Compare. If ≤ If the node's influence is insufficient, the backtracking process is terminated; the current candidate node... There are no more upstream related nodes, so the backtracking process ends. All candidate nodes that were not eliminated are aggregated into a root cause node set.

[0061] The anti-interference rule base includes: interference compensation rule set, physical constraint clauses, and historical confidence level labels;

[0062] The interference compensation rule set is a dynamic mapping relationship between the interference condition characteristics and the compensation rules established based on the root cause node.

[0063] The physical constraint clauses are the device physical boundary constraints built into each rule;

[0064] The historical confidence label is the confidence value associated with the historical success rate of each rule.

[0065] Specifically, the interference compensation rule set is used to establish a dynamic mapping relationship between the root cause nodes and the compensation strategy. The root cause nodes are stored in a graph structure, and each node is associated with an interference feature vector. ,in, For interference feature vectors, Let i represent the interference intensity, duration, and spatial distribution. It is a set of vectors consisting of n real number components. A compensation strategy is generated by matching rules from the anti-interference rule base using a dynamic mapping function.

[0066] Physical constraints are safety constraints embedded in the physical boundaries of a device, including displacement constraints, mechanical constraints, frequency domain constraints, and pressure constraints; displacement constraints The boundary value is Mechanical constraints , =150Nm; Frequency domain constraint , The inherent frequency of the device, It is a floating value. =5Hz; Pressure constraint , =0.7MPa.

[0067] Calculate the base success rate based on the historical execution records of the rules:

[0068] ,

[0069] in, Based on the success rate, The number of times to successfully eliminate interference. This represents the total number of triggers. A time decay factor is added to reduce the weight of outdated data. , The attenuation coefficient is 0.01 / day. t is the time (in days) since the most recent success. Calculate the final confidence level: .

[0070] Receive the node set output by the recursive backtracking module Extract node interference features, such as The vibration transmission coefficient and duration are determined; the compensation strategy is initialized based on expert rule data, and physical constraint clauses are embedded for each rule and associated with historical confidence labels.

[0071] S2: Based on the anti-interference rule base and multimodal data, a decision vector is generated through the causal entropy balance target of the neural symbolic reasoning engine; the decision vector is mapped to the discrete event model, and the temporal safety is verified by linear temporal logic reduction. If the verification fails, the safety policy is activated to correct the decision.

[0072] The decision vector includes: destacking height adjustment amount, vibration anti-phase control spectrum, and safety margin indicator;

[0073] The stacking height adjustment amount is a Z-axis travel compensation value generated by the neural symbolic reasoning engine based on the anti-interference rule base;

[0074] The vibration anti-phase control spectrum is a set of frequency domain waveform parameters used to actively cancel vibration interference;

[0075] The safety margin indicator is a scalar value that identifies the safety boundary of a decision.

[0076] Specifically, by matching the interference feature vectors with high-confidence rule conditions in the anti-interference rule base in real time, a rule-driven symbolic stream is output; inputting multimodal real-time data, a deep residual network is used to fit complex nonlinear relationships not covered by the rules, and a data-driven neural stream is output. Weights are assigned to the two streams:

[0077] ,

[0078] in, For dual-stream weighting, This is the current causal entropy value. The entropy balance target threshold is 0.5, and ks is the sensitivity adjustment coefficient, controlling the steepness of the function curve, with a value range of [2, 10]. When When it approaches 1, it represents a high causal entropy value dominated by symbolic flow. When it approaches 0, it represents a low causal entropy value dominated by neural flow. The current formula for calculating the causal entropy value is:

[0079] ,

[0080] in, Let be the probability of a node in the fault propagation topology graph contributing to the current anomaly, and satisfy . ; Let ln(i) be the amount of fault information at node i, with the natural logarithm ln(i) as the base. .

[0081] Synthesized output decision vector:

[0082] ,

[0083] in, Let be the decision vector. For symbolic streams, For neural flow; It comprises three core components: extracting high-confidence compensation values ​​from an anti-interference rule base and calculating the compensation amount in real time through a deep residual network, used to dynamically correct the robot's Z-axis gripping position and adjust the stacking height to compensate for the height deviation of the board material caused by interference; symbolic flow provides reference parameters and neural flow provides dynamic optimization, used to generate a vibration anti-phase control spectrum of waveform parameter set to offset vibration interference; and safety factor is calculated based on physical boundaries, and the probability of potential conflicts is predicted through a convolutional network, used as a safety margin indicator to quantify the safety risk of decision-making.

[0084] The safety strategy includes: reordering the decision action sequence that violates the linear temporal logic specification, and inserting safety checkpoints in the gaps between the reordered actions; when a safety checkpoint is triggered, compressing the stacking height adjustment amount in the decision vector to the allowable range of the safety margin, and modifying the vibration antiphase control spectrum parameters to avoid the resonance frequency band.

[0085] Specifically, the decision vector is transformed into a discrete event sequence: height adjustment event, vibration control event, and safety verification event. Formal verification is performed by adding action timing constraints, safety boundary constraints, and cooperative avoidance constraints. Action timing constraints: After vibration control is initiated, height adjustment must be performed within 500 milliseconds. Safety boundary constraints: When the safety margin indicator value is below 0.8, the height adjustment event is prohibited. Cooperative avoidance constraints: When the overlap of the working areas exceeds 40%, a reverse offset event must be triggered before subsequent actions can be executed. Verification is successful only if all specification requirements are met.

[0086] When verification fails, a three-level correction mechanism is triggered; Level 1 correction: adjust the execution order of actions and insert safety verification points at critical nodes; Level 2 correction: proportionally compress the unpacking height adjustment to the allowable safety margin range and modify the vibration anti-phase control spectrum parameters to avoid the resonance frequency band; Level 3 correction: replace the high-speed grasping action with low-speed translation and replace continuous trajectory motion with single-point positioning; the decision vector after correction must be re-verified.

[0087] S3: Monitor the actual causal entropy value generated by the decision vector and compare it with the causal entropy balance target. When the actual causal entropy value deviates from the tolerance value of the causal entropy balance target, it is marked as a high-risk signal.

[0088] Specifically, the tolerance value is calculated based on historical confidence labels and expert rule data:

[0089] ,

[0090] in, This is the tolerance value. A base tolerance of 0.1 is derived from historical success rates. This represents the historical confidence average of relevant rules in the rule base, ranging from [0, 1]. The actual entropy value is allowed to be within... Fluctuations within the range are considered significant deviations if they exceed the specified range.

[0091] Real-time calculation of actual entropy value With target value absolute deviation ,when > When a significant deviation is identified, the system automatically generates a high-risk signal, including: deviation type; relevant decision vector components; and real-time causal entropy value. and deviation ; Associated multimodal data snapshots.

[0092] S4: Based on the backtracking analysis of the root cause triggered by the high-risk signal, if the backtracking result confirms that it is a rule defect, the relevant rule set in the rule base is updated; if the backtracking result is not a rule defect, the compensation strategy is activated according to the fault type.

[0093] The root cause backtracking analysis includes: reconstructing the causal field by fusing the latest spatiotemporal interference and physical coupling parameters to form a dynamically evolving causal transmission network; traversing backward along the reconstructed causal transmission network, filtering effective paths by combining historical confidence labels in the rule base, and performing secondary verification of the physical boundary conditions of the retained nodes; if new root cause nodes not covered by the rule base or low-confidence rules are found, they are judged as rule defects; if all nodes pass the verification and conform to the rule base logic, they are judged as non-rule defects and a specific fault type classification is output.

[0094] Specifically, upon receiving a high-risk signal, a causal field reconstruction is triggered, and the causal field is updated using a tensor fusion algorithm:

[0095] ,

[0096] in, For real-time spatiotemporal interference tensors, To generate a dynamically evolving causal transmission network, tensors are physically coupled in real time. Taking the current anomalous event as the starting point for tracing the source, high-confidence paths in the rule base are prioritized, and inefficient paths with a success rate of less than 60% within the past three months are automatically filtered out; low-reliability branches with a confidence level of less than 0.7 are eliminated, and weakly influential paths with a propagation strength of less than 0.3 are removed; key nodes are retained to form a simplified topology network.

[0097] For the retained nodes, perform secondary verification of the physical boundary conditions of the equipment. The joint torque must be less than or equal to the design limit value; the vibration frequency should avoid the resonance range; the height adjustment should be strictly controlled within the ±50mm stroke range; the air source pressure fluctuation rate should be ≤0.5MPa / s; and the vacuum adsorption pressure should be maintained >85kPa.

[0098] A rule is deemed defective if any of the following conditions are met: an anomaly source not registered in the rule base; carrying a new type of interference feature vector; historical success rate of associated rules <65%; more than 10 triggers within three months with a success rate of 0%; rule confidence showing a continuous downward trend with a monthly decrease of >15%; the compensation amount recommended by the rule exceeds the device's tolerance limit; or the suggested action violates safety procedures.

[0099] When rule defects are excluded, four types of standard faults are output: gas source fluctuation fault, mechanical resonance fault, spatial interference fault, and timing out-of-synchronization fault.

[0100] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages:

[0101] This application utilizes recursive backtracking causal chain technology to accurately locate the root cause nodes in multimodal data, effectively eliminating redundant intermediate variables and significantly improving the efficiency of interference source identification. The dynamic weight allocation mechanism of the neural symbolic inference engine achieves a balance between the advantages of rule constraints and data fitting. Through the temporal security verification mechanism of linear temporal logic reduction, the risk of action sequence conflict is intercepted, and the decision security boundary is not breached by combining a three-level security strategy correction. The real-time causal entropy monitoring module accurately marks high-risk signals through dynamic tolerance thresholds. The root cause backtracking analysis adopts dynamic reconstruction technology of causal field, combined with secondary verification of physical boundaries, to achieve a rule defect judgment accuracy of 95% and shorten the fault recovery time by 60%. The collaborative causal field transforms dual-indicator compliance cases into fault-tolerant rules through a reverse growth aggregation algorithm, driving the rule base to continuously self-evolve.

[0102] Example 2: In Example 1, the root cause node was accurately located by recursively backtracking the causal chain to build an anti-interference rule base. The decision vector was generated by combining the neural symbolic reasoning engine. The adaptive adjustment of the single-machine destacking height was achieved by real-time monitoring of causal entropy and backtracking analysis triggered by high-risk signals. However, there are still shortcomings in multi-machine collaborative scenarios. This example further improves Example 1.

[0103] The causal field also includes a collaborative causal field for multi-device collaborative control: a collaborative causal field containing a causal transmission topology is constructed based on spatiotemporal interference parameters and physical coupling parameters; by recursively backtracking the chain of abnormal events of multiple devices, a minimum complete set of interference rules is extracted from the collaborative causal field and continuously updated, and this set of rules is embedded as a sub-module into the anti-interference rule base; when the collaborative causal field detects that the anomaly occurrence rate is greater than the anomaly threshold, the causal transmission topology is reconstructed in real time, the reconstructed path is injected into the neural symbolic inference engine, and collaborative control instructions are generated by fusion weight balancing and used as a component of the decision vector.

[0104] Specifically, a co-causal field is constructed based on spatiotemporal interference parameters and physical coupling parameters:

[0105] ,

[0106] in, To coordinate the causal field tensor, For cooperative state tensors.

[0107] When an anomaly in device coordination is detected, the system traces backward along the causal propagation topology. First, it locates the current abnormal event node. Second, it recursively scans upstream related nodes to verify physical feasibility. Finally, it outputs an abnormal event chain containing the sequence of key event nodes and their confidence levels. Core interference patterns are extracted from the event chain to form triple rules, which include condition vectors, interference types, and handling strategies. These rules are required to cover over 99.2% of historical abnormal event scenarios. Once this requirement is met, the rules are dynamically embedded into an anti-interference rule base, forming a positive-interactive complementary structure with existing rules.

[0108] When the real-time monitoring shows an anomaly occurrence rate > ( Reconstruction is initiated when the risk level is 0.85, and the topology edge weights are dynamically adjusted based on the risk level.

[0109] ,

[0110] in, For the new dual-flow weight, The original dual-stream weights are denoted by Δt, which is the time interval since the last anomaly, and kt is the time decay coefficient, with a value range of [0.02, 0.12]. The normalized risk coefficient output by the spiking neural network, with a value range of [0, 1], is used; the differential manifold generating the optimized propagation path is injected into the input layer of the neural symbolic inference engine.

[0111] ,

[0112] in, For cooperative control vectors, For the rule base decision vector, To reconstruct the path manifold, A weight of 0.4 corresponds to a knowledge-experience-driven setting. The output is a three-dimensional collaborative control vector containing XYZ axis compensation vectors, clock synchronization correction coefficients, and load dynamic distribution matrix.

[0113] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages:

[0114] This application addresses the problem of adaptive adjustment of destacking height in multi-machine collaborative scenarios by constructing a collaborative causal field: a collaborative causal field tensor is generated based on spatiotemporal interference parameters and physical coupling parameters. When an abnormality in equipment collaboration is detected, an abnormal event chain is generated by recursively backtracking along the causal transmission topology. The minimum complete set of interference rules is extracted from the chain and dynamically embedded into the anti-interference rule base. When the abnormality rate exceeds a threshold, the causal transmission topology is reconstructed in real time, and the optimized path is injected into the neural symbolic inference engine. Through the fusion of weight balance, collaborative control commands including spatial compensation, beat synchronization, and load distribution are generated to achieve highly collaborative adjustment of multiple devices in scenarios of vibration interference and spatial conflict.

[0115] Example 3: Example 2 solves the problem of adaptive adjustment of destacking height in multi-machine collaborative scenarios by constructing a collaborative causal field, but it lacks an active cancellation mechanism for the vibration spectrum of the faulty machine and the dynamic optimization capability of the rule base is insufficient. This example further improves upon Example 2.

[0116] The collaborative causal field also includes: constructing a collaborative fault-tolerant control field based on multimodal data; synchronously executing the acquisition of vibration spectrum of the faulty machine to generate an anti-phase wave for cancellation by the neighboring machine in the collaborative fault-tolerant control field; and monitoring the reverse offset triggered when the overlap of the working area exceeds 40%; classifying the collaborative handling of vibration interference and spatial conflict; verifying whether the results simultaneously meet the dual requirements of a production capacity recovery rate greater than 95% and an interference elimination degree greater than 90%; and updating the fault-tolerant rules of the qualified cases to the anti-interference rule base by generating fault-tolerant rules through the reverse growth aggregation algorithm.

[0117] Specifically, the vibration spectrum of the faulty machine is collected in real time to generate an anti-phase wave, which is synchronously output by the actuator of the adjacent machine to cancel the vibration energy. The anti-phase wave is sent to the adjacent machine through a low-latency communication network, and is output in real time by the vibration suppressor of the adjacent machine to achieve active cancellation of vibration energy. The lidar matrix monitors the overlap of the working area. When the overlap rate is >40%, reverse offset control is triggered to dynamically adjust the pose in the opposite direction of the collision risk to eliminate the potential for spatial conflict. Vibration interference and spatial conflict are handled in a hierarchical and collaborative manner: Level 1 problem (vibration interference): activate anti-phase cancellation with a priority weight of 70%; Level 2 problem (spatial conflict): trigger reverse offset with a priority weight of 90%; Composite problem (vibration + space): priority weight is 1:1.3, with two strategies in parallel. The dual hard indicators of production capacity recovery rate ≥95% and interference elimination degree ≥90% are strictly verified. The characteristic parameters (spectral characteristics, offset, weight ratio) of the qualified cases are clustered into fault-tolerant rules through a reverse growth aggregation algorithm, embedded into the anti-interference rule base with an initial confidence of 80%, and the confidence is dynamically adjusted according to the subsequent execution success rate.

[0118] Based on the aforementioned patented method, this application also provides an adaptive adjustment system for the depalletizing height of a stamping robot, such as... Figure 2 As shown, the system includes:

[0119] Multimodal data acquisition module: Real-time acquisition of equipment body parameters, process characteristic parameters, spatiotemporal interference parameters, physical coupling parameters, collaborative state parameters and expert rule data;

[0120] Recursive backtracking analysis module: Constructs a fault propagation topology graph based on multimodal data, integrates spatiotemporal interference and physical coupling parameters to generate a causal field, and traverses the causal propagation topology manifold in reverse to extract the set of root cause nodes;

[0121] Rule base construction module: Generates an anti-interference rule base based on the root cause node, which includes interference compensation rule set, physical constraint clauses and historical confidence labels;

[0122] The neural symbolic decision-making module generates decision vectors through the neural symbolic reasoning engine, performs linear temporal logic reduction verification, and activates security policy correction decisions when verification fails.

[0123] Causal Entropy Monitoring Module: Monitors the actual causal entropy value generated by the decision vector and marks a high-risk signal when it deviates from the target tolerance value;

[0124] Cooperative causal field processing module: Constructs a cooperative causal field based on spatiotemporal interference and physical coupling parameters, extracts the minimum complete interference rule set and embeds it into the rule base, and reconstructs the topology path to generate cooperative control instructions when anomalies exceed the limit;

[0125] Fault-tolerant control execution module: Synchronously executes vibration anti-phase cancellation and overlap-triggered reverse offset in the collaborative fault-tolerant control field, and handles collaborative issues in a hierarchical manner through the resource rescheduling engine;

[0126] Closed-loop verification module: Verifies the achievement of dual targets of capacity recovery rate ≥95% and interference elimination degree ≥90%;

[0127] Rule self-evolution module: Transforms compliant cases into fault-tolerant rules through a reverse growth aggregation algorithm and updates the anti-interference rule base in real time.

[0128] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages:

[0129] This application addresses the issues of vibration transmission and dynamic spatial conflict between devices in multi-machine collaborative scenarios. It constructs a fault-tolerant control mechanism within a collaborative causal field framework: a reverse cancellation signal is generated by real-time acquisition of the vibration spectrum of the faulty machine, which is synchronously output by the actuators of neighboring machines to block vibration energy transmission; simultaneously, the overlapping state of the working areas is monitored, and adaptive pose offset control is triggered when a safety threshold is exceeded to eliminate collision risks; hierarchical collaborative handling is implemented for both vibration and spatial interference problems, dynamically allocating handling weights according to problem type, and employing a parallel execution strategy in complex fault scenarios; a dual-indicator verification system centered on capacity recovery and interference elimination is established to ensure that the handling effect meets production requirements; finally, the characteristic patterns of successful cases are transformed into fault-tolerant rules through an intelligent aggregation algorithm, embedded into the rule base, and a confidence dynamic adjustment mechanism is established to form a closed-loop self-optimization capability.

[0130] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for adaptive adjustment of the destacking height of a stamping robot, characterized in that, include: S1: Acquire multimodal data, extract the root cause nodes of the multimodal data by recursively backtracking the causal chain, and generate an anti-interference rule base based on the interference condition characteristics and compensation strategy mapping relationship of the root cause nodes. S2: Based on the anti-interference rule base and multimodal data, a decision vector is generated through the causal entropy balance objective of the neural symbolic reasoning engine; The decision vector is mapped to a discrete event model, and the timing safety is verified by linear temporal logic reduction. If the verification fails, the safety policy is activated to correct the decision. S3: Monitor the actual causal entropy value generated by the decision vector and compare it with the causal entropy balance target. When the actual causal entropy value deviates from the tolerance value of the causal entropy balance target, it is marked as a high-risk signal. S4: Based on the backtracking analysis of the root cause triggered by the high-risk signal, if the backtracking result confirms that it is a rule defect, the relevant rule set in the rule base is updated; if the backtracking result is not a rule defect, the compensation strategy is activated according to the fault type.

2. The adaptive adjustment method for the destacking height of a stamping robot as described in claim 1, characterized in that, The multimodal data includes: equipment body parameters, process characteristic parameters, spatiotemporal interference parameters, physical coupling parameters, cooperative state parameters, and expert rule data; The equipment body parameters are the real-time status parameters of joint torque and motor current recorded in the equipment operation log. The process characteristic parameters are the process parameters of die size and stamping speed in the stamping process database; The spatiotemporal interference parameters are the workspace overlap and clock synchronization deviation rate collected by the sensor network. The physical coupling parameters are process parameter coupling data that include the vibration transmission coefficient and the gas source fluctuation coefficient; The collaborative status parameters are the local fault codes, neighboring machine operating status, and overall production capacity margin collected by the collaborative sensing nodes. The expert rule data refers to historical fault handling rules provided by expert experience documents.

3. The adaptive adjustment method for the destacking height of a stamping robot as described in claim 1, characterized in that, The recursive backtracking causal chain includes: starting from the current abnormal event, constructing a fault propagation topology graph based on multimodal data; fusing spatiotemporal interference parameters and physical coupling parameters through the fault propagation topology graph to generate a causal field; performing a reverse traversal along the causal transmission topology manifold in the causal field; and deleting unnecessary intermediate variables based on the confidence level of historical handling cases; performing equipment physical boundary condition verification on candidate nodes; performing a removal operation on nodes that exceed the physical boundary; stopping the backtracking when the node's influence is below the threshold defined in the expert experience document or reaches the initial boundary node of the process system; and outputting a set of root cause nodes.

4. The adaptive adjustment method for the destacking height of a stamping robot as described in claim 1, characterized in that, The anti-interference rule base includes: interference compensation rule set, physical constraint clauses, and historical confidence level labels; The interference compensation rule set is a dynamic mapping relationship between the interference condition characteristics and the compensation rules established based on the root cause node. The physical constraint clauses are the device physical boundary constraints built into each rule; The historical confidence label is the confidence value associated with the historical success rate of each rule.

5. The adaptive adjustment method for the destacking height of a stamping robot as described in claim 1, characterized in that, The decision vector includes: destacking height adjustment amount, vibration anti-phase control spectrum, and safety margin indicator; The stacking height adjustment amount is a Z-axis travel compensation value generated by the neural symbolic reasoning engine based on the anti-interference rule base; The vibration anti-phase control spectrum is a set of frequency domain waveform parameters used to actively cancel vibration interference; The safety margin indicator is a scalar value that identifies the safety boundary of a decision.

6. The adaptive adjustment method for the destacking height of a stamping robot as described in claim 1, characterized in that, The safety strategy includes: reordering the decision action sequence that violates the linear temporal logic specification, and inserting safety checkpoints in the gaps between the reordered actions; when a safety checkpoint is triggered, compressing the stacking height adjustment amount in the decision vector to the allowable range of the safety margin, and modifying the vibration antiphase control spectrum parameters to avoid the resonance frequency band.

7. The adaptive adjustment method for the destacking height of a stamping robot as described in claim 1, characterized in that, The root cause backtracking analysis includes: reconstructing the causal field by fusing the latest spatiotemporal interference and physical coupling parameters to form a dynamically evolving causal transmission network; traversing backward along the reconstructed causal transmission network, filtering effective paths by combining historical confidence labels in the rule base, and performing secondary verification of the physical boundary conditions of the retained nodes; if new root cause nodes not covered by the rule base or low-confidence rules are found, they are judged as rule defects; if all nodes pass the verification and conform to the rule base logic, they are judged as non-rule defects and a specific fault type classification is output.

8. The adaptive adjustment method for the destacking height of a stamping robot as described in claim 3, characterized in that, The causal field also includes a collaborative causal field for multi-device collaborative control: a collaborative causal field containing a causal transmission topological manifold is constructed based on spatiotemporal interference parameters and physical coupling parameters; by recursively backtracking the chain of abnormal events of multiple devices, a minimum complete set of interference rules is extracted from the collaborative causal field and continuously updated, and this set of rules is embedded as a sub-module into the anti-interference rule base. When the co-causal field detects an anomaly rate greater than the anomaly threshold, it reconstructs the causal transmission topology in real time, injects the reconstructed path into the neural symbolic inference engine, generates co-control instructions through weight balancing, and uses them as a component of the decision vector.

9. The adaptive adjustment method for the destacking height of a stamping robot as described in claim 8, characterized in that, The collaborative causal field also includes: constructing a collaborative fault-tolerant control field based on multimodal data; synchronously executing the acquisition of vibration spectrum of the faulty machine to generate an anti-phase wave for cancellation by the neighboring machine in the collaborative fault-tolerant control field; and monitoring the reverse offset triggered when the overlap of the working area exceeds 40%; classifying the collaborative handling of vibration interference and spatial conflict; verifying whether the results simultaneously meet the dual requirements of a production capacity recovery rate greater than 95% and an interference elimination degree greater than 90%; and updating the fault-tolerant rules of the qualified cases to the anti-interference rule base by generating fault-tolerant rules through the reverse growth aggregation algorithm.

10. A stamping robot depalletizing height adaptive adjustment system, applied to the stamping robot depalletizing height adaptive adjustment method as described in any one of claims 1 to 9, characterized in that, The system includes: Multimodal data acquisition module: Real-time acquisition of equipment body parameters, process characteristic parameters, spatiotemporal interference parameters, physical coupling parameters, collaborative state parameters and expert rule data; Recursive backtracking analysis module: Constructs a fault propagation topology graph based on multimodal data, integrates spatiotemporal interference and physical coupling parameters to generate a causal field, and traverses the causal propagation topology manifold in reverse to extract the set of root cause nodes; Rule base construction module: Generates an anti-interference rule base based on the root cause node, which includes interference compensation rule set, physical constraint clauses and historical confidence labels; The neural symbolic decision-making module generates decision vectors through the neural symbolic reasoning engine, performs linear temporal logic reduction verification, and activates security policy correction decisions when verification fails. Causal Entropy Monitoring Module: Monitors the actual causal entropy value generated by the decision vector and marks a high-risk signal when it deviates from the target tolerance value; Cooperative causal field processing module: Constructs a cooperative causal field based on spatiotemporal interference and physical coupling parameters, extracts the minimum complete interference rule set and embeds it into the rule base, and reconstructs the topology path to generate cooperative control instructions when anomalies exceed the limit; Fault-tolerant control execution module: Synchronously executes vibration anti-phase cancellation and overlap-triggered reverse offset in the collaborative fault-tolerant control field, and handles collaborative issues in a hierarchical manner through the resource rescheduling engine; Closed-loop verification module: Verifies the achievement of dual targets of capacity recovery rate ≥95% and interference elimination degree ≥90%; Rule self-evolution module: Transforms compliant cases into fault-tolerant rules through a reverse growth aggregation algorithm and updates the anti-interference rule base in real time.

Images