Injection mold action timing deviation detection method based on state sequence semantic coding and server

Abstract

The application discloses a kind of based on state sequence semantic coding injection mold action timing deviation detection method and server, method includes: first, the action state information sequence of target injection mold in complete forming cycle is acquired, the action state information unit in sequence is reorganized according to semantic association mold forming process logic, process semantic topological graph is generated and mapped to high-dimensional manifold space, global action semantic manifold structure is constructed and calculates homology group feature;Second, the reference homology group feature of reference action semantic manifold structure of historical non-deviation period is acquired to generate global configuration deviation detection result including survival interval distribution deviation and ring structure collapse degree parameter Homology group disturbance analysis;Finally, according to global configuration deviation detection result, the deviation position and deviation propagation path of action timing are determined and correction instruction is generated, can identify the starting point, influence scope and propagation direction of action timing anomaly from global topological configuration level.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a method and server for detecting timing deviations in injection mold actions based on state sequence semantic encoding. Background Technology

[0002] In the injection molding process, the timing of mold movements directly determines the quality and production efficiency of the molded product. Traditional injection mold movement monitoring methods typically rely on threshold judgments of individual sensor signals or simple action sequences. For example, they determine whether the action is executed normally by detecting whether the displacement sensor's stroke is complete or whether the pressure sensor's peak value meets the standard.

[0003] With the development of industrial automation and intelligent manufacturing technologies, some existing technologies have begun to attempt joint analysis of multiple action state parameters of injection molds. For example, they record complete action sequences such as mold closing, injection, pressure holding, mold opening, and ejection, and identify anomalies by comparing them with preset standard timing templates. However, most of the above-mentioned existing methods focus on whether the execution of individual actions is timely or in place, and it is difficult to depict the complex dependencies and collaborative logic between various process units of the mold as a whole.

[0004] Therefore, the problem with the existing technology is that when the timing of the mold action deviates, it is impossible to identify the starting position of the abnormality from the perspective of global collaborative configuration, assess its impact on subsequent processes, and track the propagation path of the deviation in the process network, which makes it difficult to achieve accurate positioning and effective correction intervention. Summary of the Invention

[0005] The purpose of this invention is to provide a method and server for detecting timing deviations in injection mold actions based on state sequence semantic encoding, so as to solve the problems mentioned in the background art.

[0006] This invention provides a method for detecting timing deviations in injection mold actions based on state sequence semantic encoding, comprising: The sequence of action status information collected sequentially over time during the complete molding cycle of the target injection mold is obtained, and the sequence of action status information contains multiple action status information units. The action state information units in the action state information sequence are semantically associated and reorganized according to the mold forming process logic to generate a process semantic topology graph containing functional semantic nodes and logical coupling relationship edges. Based on the functional semantic nodes and logical coupling relationship edges of the process semantic topology graph, the process semantic topology graph is mapped to a high-dimensional manifold space to construct the global action semantic manifold structure corresponding to the complete forming cycle, and the homology group feature of the global action semantic manifold structure in the homology group dimension is calculated. The reference action semantic manifold structure of the target injection mold is pre-constructed within the historical non-deviation molding cycle. The reference homology group features of the reference action semantic manifold structure are extracted. Based on the homology group features of the global action semantic manifold structure and the reference homology group features, homology group perturbation analysis is performed to generate global configuration deviation detection results. The global configuration deviation detection results include the survival interval distribution offset of the cooperation relationship between functional semantic nodes in the process semantic topology graph and the ring structure collapse degree parameter. Based on the global configuration deviation detection results, the deviation position and deviation propagation path of the target injection mold in the complete molding cycle are determined, and a mold action timing correction command is generated based on the deviation position and deviation propagation path.

[0007] This invention provides a timing deviation detection server for injection mold actions, comprising: A processor; a storage device on which a computer program is stored; a network interface for providing network communication functions; when the computer program is executed by the processor, the processor implements the above-described injection mold action timing deviation detection method based on state sequence semantic encoding.

[0008] The present invention provides a readable storage medium on which a program or instruction is stored, and when the program or instruction is executed by a processor, it implements the above-described injection mold action timing deviation detection method based on state sequence semantic encoding.

[0009] Compared with existing technologies, the beneficial effects of this invention are as follows: By semantically associating and recombining the sequence of action state information continuously collected during the injection mold molding cycle according to the molding process logic, a process semantic topology graph capable of depicting the inter-process dependencies is generated. This process semantic topology graph is then mapped to a high-dimensional manifold space to construct a global action semantic manifold structure, and its homology group features are calculated, achieving a topological invariant description of the mold action collaboration logic. This allows complex action collaboration relationships to be quantified into stable algebraic features. By introducing a pre-constructed baseline action semantic manifold structure and its baseline homology group features from historical undevised cycles, and performing perturbation analysis with the homology group features of the current cycle, a global configuration deviation detection result containing parameters such as survival interval distribution offset and annular structure collapse degree can be generated. This accurately captures abnormal changes in the collaboration relationships between functional semantic nodes in the process semantic topology graph. Finally, based on the deviation detection results, the deviation position and deviation propagation path of the action sequence are determined, and correction instructions are generated. This achieves the identification of the starting point, influence range, and propagation direction of action sequence anomalies from the global topology configuration level, thereby realizing precise control and intelligent operation and maintenance of the injection mold. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the drawings used in the description of the embodiments or related technologies 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.

[0011] Figure 1 This is a flowchart of a method for detecting timing deviations in injection mold actions based on state sequence semantic encoding, provided in an embodiment of the present invention.

[0012] Figure 2 This is a schematic diagram of a method for detecting timing deviations in injection mold actions based on state sequence semantic encoding, provided in an embodiment of the present invention.

[0013] Figure 3 This is a schematic diagram of the basic structure of a timing deviation detection server for injection mold actions provided in an embodiment of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] To facilitate understanding of the entire technical solution, the relevant technical terms will first be explained by example below: The action state information sequence is a sequence of multiple action state information units collected continuously in chronological order, recording the state changes of the mold throughout the entire forming cycle.

[0016] Functional semantic nodes are formed by fusing adjacent action state information units with direct driving and transmission relationships, corresponding to a complete process transmission semantic unit.

[0017] Logical coupling edges are used to characterize the directed connections between different functional semantic nodes, representing the strength of process dependencies and action connection constraints.

[0018] The process semantic topology graph is a graph structure composed of functional semantic nodes and logical coupling edges, which describes the logical relationships of the mold forming process.

[0019] High-dimensional manifold space: is a continuous geometric space in which the semantic topology of the process is mapped, so that the logical relationship between nodes is reflected by spatial distance.

[0020] The global action semantic manifold structure is a topology-preserving distribution configuration of action semantic nodes in a high-dimensional manifold space, consisting of manifold embedding coordinates and neighborhood connectivity.

[0021] Homology group features are used to describe the algebraic invariants of the number of connected holes and the distribution of ring structures in the global action semantic manifold structure.

[0022] The baseline action semantic manifold structure is a standard action semantic manifold structure that is pre-constructed within a historical, unbiased forming cycle and represents the normal state.

[0023] The baseline homology group features are homology group features extracted from the semantic manifold structure of the baseline action, and serve as a reference benchmark for judging whether the current cycle has deviated.

[0024] Homology group perturbation analysis and processing: The analysis process of comparing the current homology group characteristics with the baseline homology group characteristics to identify the perturbation type and magnitude.

[0025] The global configuration deviation detection results include the detection results of survival interval distribution offset and ring structure collapse degree parameters, which are used to quantify the degree of topological deviation of action timing.

[0026] The survival interval distribution offset refers to the degree of change in the length of the interval in the manifold space where logically coupled edges persist, relative to the baseline.

[0027] The parameter representing the degree of collapse of the ring structure is a quantitative indicator of the degradation of the integrity of closed dependency loops formed by functional semantic nodes in the global configuration.

[0028] Deviation Position: The time position range and corresponding functional semantic node of the action sequence deviation in the complete forming cycle.

[0029] Deviation from propagation path: The sequence of functional semantic nodes affected by the propagation along the directed direction of the logical coupling relationship, starting from the deviated initial functional semantic node.

[0030] The mold action timing correction instruction includes the identifier of the functional semantic node to be corrected, the identifier of the target timing interval after correction, and the instruction to adjust the action execution order.

[0031] Based on the above, please refer to Figure 1 , Figure 1 This is a flowchart of a method for detecting timing deviations in injection mold actions based on state sequence semantic encoding, provided by an embodiment of the present invention. The method can be executed by an injection mold action timing deviation detection server, or by both the injection mold action timing deviation detection server and the server. The method includes steps 110-150.

[0032] This invention enables precise positioning and effective corrective intervention when timing deviations occur in mold operations. From a global collaborative configuration perspective, it identifies the starting point of the anomaly, assesses its impact on subsequent processes, and tracks the propagation path of the deviation within the process network. In this embodiment, the target injection mold is a horizontal injection molding machine equipped with a multi-channel sensor data acquisition system. Its complete molding cycle includes multiple continuously executed process stages such as mold closing, injection, pressure holding, plasticizing, mold opening, and ejection.

[0033] Step 110: Obtain the sequence of action status information collected continuously in time order during the complete molding cycle of the target injection mold, wherein the sequence of action status information contains multiple action status information units.

[0034] Specifically, a real-time data acquisition channel is established between the target injection mold control system and a sensor network deployed on key moving parts of the mold via an industrial Ethernet protocol. The sensor network includes at least a displacement sensor mounted on the moving platen, a speed sensor mounted on the ejector pin, a pressure sensor mounted on the injection screw, and a temperature sensor mounted on the mold cavity wall. A fixed sampling time interval is set, which is predetermined based on the dynamic response characteristics of the injection molding process.

[0035] At each sampling moment, all sensors in the sensor network synchronously acquire data, and each sensor outputs a raw measurement value. The raw measurement values ​​from all sensors at the same sampling moment are concatenated according to a preset dimensional order to form a fixed-length feature vector, which is a motion state information unit. Each motion state information unit includes at least the moving platen position encoding value, the ejector rod movement speed encoding value, the cavity pressure encoding value, and the mold characteristic point temperature encoding value.

[0036] All action state information units acquired at all sampling times are arranged in ascending order of time to form an action state information sequence. The length of the sequence is equal to the total duration of the complete forming cycle divided by the sampling time interval. Each element in the sequence corresponds to a unique time index, which implicitly represents the relative time position of the action state information unit in the forming cycle.

[0037] Step 120: Semantically associate and reorganize the action state information units in the action state information sequence according to the mold forming process logic to generate a process semantic topology graph containing functional semantic nodes and logical coupling relationship edges.

[0038] It is understandable that the core of step 120 is to abstract macroscopic action segments with clear process meanings from the original high-frequency sampling data, reveal the logical dependencies between the above segments, and convert the time series data into graph structure data.

[0039] Step 121: Extract the action timing position identifier and action type label of each action status information unit in the action status information sequence, and determine the time sequence relationship and action category attribute of each action status information unit within the complete forming cycle.

[0040] Iterate through each action state information unit in the action state information sequence. For the i-th unit, directly use its position index i in the sequence as its action temporal position identifier. Input the feature vector of this unit into a pre-trained action classifier, which adopts a multilayer perceptron architecture, containing one input layer, two hidden layers, and one output layer. The number of neurons in the input layer is equal to the dimension of the action state information unit.

[0041] The first hidden layer uses a modified linear unit (MRU) as the activation function to perform a non-linear transformation on the input features. The second hidden layer also uses a MRU to further extract high-level semantic features. The output layer uses a flexible maximum function to output the probability distribution of the feature vector belonging to each action type in a predefined set of action types.

[0042] The predefined set of action types includes: low-speed mold closing forward, high-pressure mold closing locking, injection seat forward, screw injection forward, pressure holding, screw metering rotation, plasticizing back pressure establishment, injection seat retraction, slow mold opening, fast mold opening, ejector rod forward, and ejector rod retraction. The action type corresponding to the maximum output probability is selected as the action type label for this action status information unit.

[0043] Thus, each action state information unit is assigned an action timing location identifier and an action type label, which together determine the precise time point of the unit within the complete forming cycle and the type of micro-action it performs.

[0044] Step 122: Based on the time sequence relationship and action category attribute, identify adjacent action state information unit pairs in the action state information sequence that satisfy the preset mold action process constraints. The preset mold action process constraints include action execution mechanism dependency constraints and action transmission direction constraints.

[0045] A sliding window scan is performed on the sequence of action state information. The window size is 2, and the step size is 1. Each pair of adjacent action state information units, i.e., the i-th unit and the (i+1)-th unit, is checked sequentially. Pre-defined mold action process constraints are a set of hard rules used to determine whether two consecutive units should be merged into the same high-level semantic unit. The action execution mechanism dependency constraint rules are as follows: A predefined execution mechanism mapping table is queried, which records the identifier of the main driving execution mechanism corresponding to each action type tag. If the execution mechanism identifier corresponding to the action type tag of the i-th unit is the same as the execution mechanism identifier corresponding to the action type tag of the (i+1)-th unit, then the mechanism dependency constraint is satisfied.

[0046] The motion transmission direction constraint rules are as follows: A predefined motion transmission direction matrix is ​​queried. This matrix is ​​a square matrix where rows and columns are composed of all motion type labels. Each element value in the matrix indicates whether the transmission direction from the motion type of the i-th unit to the motion type of the (i+1)-th unit is physically feasible, then the direction constraint is satisfied. Only when adjacent unit pairs simultaneously satisfy both the motion actuator dependency constraint and the motion transmission direction constraint are these unit pairs marked as candidate fusion pairs that satisfy the preset mold motion process constraints.

[0047] The action classifier used in step 122 has a multilayer perceptron architecture consisting of one input layer, two hidden layers, and one output layer. The number of neurons in the input layer is equal to the dimension of the action state information unit. The first hidden layer contains 128 neurons, using a modified linear unit as the activation function to perform a max(0, x) nonlinear transformation on the input of this layer. The second hidden layer contains 64 neurons, also using a modified linear unit. The output layer contains a number of neurons equal to the size of the predefined action type set, using a flexible maximum activation function to convert the output into a probability distribution. The training process of this classifier is as follows: The training data comes from historically collected sequence of injection mold action state information, from which a total of 50,000 labeled action type sample units are extracted. Each sample unit uses its corresponding feature vector as input and its correct action type as the label. The cross-entropy loss function is used to calculate the difference between the predicted probability distribution and the true label. The Adam optimizer was used for parameter updates, with an initial learning rate of 0.001, an exponential decay rate beta1 of 0.9 for the first moment estimation, and an exponential decay rate beta2 of 0.999 for the second moment estimation. The batch size was set to 32, and the training run consisted of 50 epochs. An early stopping strategy was employed, stopping training when the validation set loss no longer decreased after 5 consecutive epochs. The evaluation metric was classification accuracy. In inference applications, for each action state information unit to be classified, its feature vector was directly input into the trained classifier. The action type corresponding to the maximum value in the probability distribution generated by the output layer was the predicted action type label for that unit.

[0048] Step 123: Semantic fusion processing is performed on adjacent action state information unit pairs that satisfy the preset mold action process constraints. The action type tags of the adjacent action information unit pairs are merged into composite action semantic tags, and the action temporal position identifiers of the adjacent action state information unit pairs are combined into composite temporal interval identifiers.

[0049] For each pair of adjacent units marked as a candidate fusion pair, a semantic fusion operation is performed. The action type tag of the i-th unit and the action type tag of the (i+1)-th unit are concatenated using a connector to generate a string, which is the composite action semantic tag.

[0050] If the action type of the i-th unit is labeled as screw injection forward and the action type of the (i+1)-th unit is labeled as pressure holding and maintenance, then the semantic label of the composite action is screw injection forward connected to pressure holding and maintenance. Using the action timing position identifier i of the i-th unit as the starting index and the action timing position identifier i+1 of the (i+1)-th unit as the ending index, a closed interval is formed, which is the composite timing interval identifier.

[0051] If multiple consecutive adjacent units are recursively merged, for example, the i-th unit, the (i+1)-th unit, and the (i+2)-th unit satisfy the constraints in pairs, then the final composite action semantic label will concatenate the three action type tags in sequence, and the composite temporal interval will be identified as the starting index i to the ending index i+2.

[0052] Step 124: Generate the functional semantic node based on the composite action semantic tag and the composite time interval identifier. The functional semantic node corresponds to a complete process transmission semantic unit in the mold action process.

[0053] The result of each semantic fusion operation in steps 123, or the final result after recursive fusion, is defined as a functional semantic node. Each functional semantic node (Node) u Each node has a unique node identifier, which encapsulates two core attributes: one is a composite action semantic label. u The first is to summarize the complete semantics of the macroscopic action represented by the node; the second is the composite temporal interval identifier Interval. u Its form is from start index to end index, which marks the time range covered by the node in the original time sequence. A functional semantic node no longer corresponds to the micro state of a single sampling moment, but corresponds to a complete and indivisible process transmission semantic unit, such as the mold closing and locking joint action unit, the injection holding pressure conversion action unit, and the mold opening and ejection linkage action unit.

[0054] Step 125: Extract the process dependency strength and action connection constraints between different functional semantic nodes under the mold forming process logic. The process dependency strength is determined according to the pre- and post-process dependencies of the corresponding process of the functional semantic node in the forming cycle. The action connection constraints are determined according to the switching order of the action execution mechanism and the action transmission time interval of the corresponding process of the functional semantic node.

[0055] After generating all functional semantic nodes, it is necessary to analyze the directed relationships between the nodes. For any two functional semantic nodes... u and Node v Interval is identified based on their composite time interval. u and Interval v Determine the sequence order. If Interval u The ending index is less than the Interval v The starting index determines the Node. u It is a Node v Preceding nodes. Dependency strength of processes. struv It is a quantitative indicator, and its calculation is based on statistical analysis of historical normal production data.

[0056] Extract a large number of action state information sequences without deviation from the forming cycle from the historical database, and construct the corresponding functional semantic node sequences according to the steps described above. For each pair of nodes with a predecessor-successor relationship... u and Node v The combination of Nodes in all historical cycles is statistically analyzed. u Then Node is executed. v Freq uv Simultaneously, statistics on Nodes across all historical periods are compiled. u Total number of times any subsequent nodes are executed Frequ Then the process dependence intensity Depend struv =Freq uv / Total Frequ The closer this value is to 1, the stronger the Node. u Execute Node after completion v The higher the probability, the better. Action connection constraints include two sub-items: actuator switching sequence constraints and action transmission time interval constraints.

[0057] Regarding the actuator switching sequence constraint, from Node u Parsing the semantic tags of the composite actions reveals the actuator identifier A corresponding to the last micro-action. end From Node v Parsing the semantic tag of the compound action reveals the actuator identifier B corresponding to the first micro-action. start If A end With B start Same or A end To B start If the switch is within the predefined set of allowed switches, the sequence is considered valid. The action transmission time interval constraint is then calculated using the standard time interval Std. gapuv Its value is a Node obtained from historical statistics. v The starting index and Node u The difference between the ending indices is multiplied by the average of the sampling time intervals.

[0058] Step 126: Generate the logical coupling relationship edge based on the process dependency strength and action connection constraint. The logical coupling relationship edge connects two functional semantic nodes with process dependency relationship and marks the connection strength and constraint type.

[0059] For each pair of functional semantic nodes with a preceding and following dependency relationship, u and Node v Generate a directed logical coupling edge. uvThe direction is from Node u Pointing to Node v In the data structure of this side, an attribute field is set to store the connection strength. The value of this field is the process dependency strength (Depend) calculated in step 125. struv .

[0060] Additionally, another attribute field is set to store the constraint type. This field is a composite data structure containing two subfields: the first subfield stores the validity flag of the actuator switching order, with a value of either valid or invalid; the second subfield stores the standard time interval Std for action transmission. gapuv Thus, each logical coupling edge not only connects two functional semantic nodes, but also carries the quantitative characteristics of their collaborative relationship.

[0061] Step 127: Combine all functional semantic nodes and all logical coupling relationship edges into a process semantic topology graph.

[0062] All generated functional semantic nodes are treated as vertices, and all generated logical coupling relationship edges are treated as directed edges, forming a directed graph structure, denoted as the process semantic topology graph. The vertex set of this graph represents all identified macroscopic process semantic units in the complete molding cycle, and the edge set represents the temporal dependencies and collaborative constraints between these process units. The process semantic topology graph is an attributed directed graph, where each vertex and each edge is accompanied by various attribute information calculated in steps 124 and 126.

[0063] Step 130: Based on the functional semantic nodes and logical coupling relationship edges of the process semantic topology graph, map the process semantic topology graph to a high-dimensional manifold space, construct the global action semantic manifold structure corresponding to the complete forming cycle, and calculate the homology group features of the global action semantic manifold structure in the homology group dimension.

[0064] This step embeds a discrete graph structure into a continuous high-dimensional geometric space, allowing the logical relationships between nodes to be represented by spatial distance, and extracting the topological invariants of the geometric structure.

[0065] Step 131: Extract the composite action semantic label and composite time interval identifier of each functional semantic node in the process semantic topology diagram, and construct the high-dimensional coordinate initialization parameters of each functional semantic node. The high-dimensional coordinate initialization parameters include action semantic dimension coordinates and time interval dimension coordinates.

[0066] For each functional semantic node in the process semantic topology graph u First, obtain its composite action semantic label. uDefine an action semantic space whose dimension is equal to the total number of possible compound action semantic labels. (The last part, "Label," appears to be a typo and can be left as is.) u This is represented as a one-hot encoded vector in the space, the length of which is equal to the total number of labels, and the vector contains elements related to the Label. u The corresponding dimension position takes a preset activation value, and all other dimension positions take preset inactive values. This one-hot encoded vector constitutes a Node. u The coordinates of the action semantic dimension.

[0067] Next, obtain its composite time-series interval identifier, Interval. u This is equal to the distance from the start index to the end index. The start and end indices are treated as two independent dimension values, forming a two-dimensional temporal interval dimension coordinate, in the form of start index value and end index value. Finally, the action semantic dimension coordinate vector and the temporal interval dimension coordinate vector are concatenated end-to-end to form a high-dimensional vector with a dimension equal to the total number of tags plus 2. This vector is the Node. u Initial coordinates in a high-dimensional manifold space.

[0068] Step 132: Extract the connection strength and constraint type of each logical coupling relationship edge in the process semantic topology graph, and construct high-dimensional connection weight parameters for the logical coupling relationship edges. The high-dimensional connection weight parameters are used to characterize the distance constraint relationship between functional semantic node pairs in the high-dimensional manifold space.

[0069] For each logical coupling edge in the process semantic topology graph uv Obtain its connection strength Depend struv and the standard time interval Std in the constraint type gapuv Construct a high-dimensional connection weight parameter. uv This parameter is a scalar whose value is determined by both the connection strength and the time interval. Design a fusion function that first converts the standard time interval Std... gapuv Multiplying by a negative adjustment coefficient and using it as the input to the exponential part of the exponential function yields a time decay factor that decreases exponentially as the time interval increases.

[0070] Then, the connection strength will be increased. struv Multiplying this by the time decay factor and then compressing the product to an open interval of 0 to 1 using a nonlinear mapping function yields the final high-dimensional connection weight parameter, Weight. uv Weight uv The closer it is to the upper limit, the better the Node is in the manifold embedding process. u and Node v They should be placed very close to each other; Weightuv The closer to the lower limit, the weaker the distance constraint between the two.

[0071] Step 133: Based on the high-dimensional coordinate initialization parameters and the high-dimensional connection weight parameters, a manifold dimensionality reduction and preservation algorithm is used to map all functional semantic nodes of the process semantic topology graph to a high-dimensional manifold space of a preset dimension.

[0072] A constrained Laplacian eigenmap algorithm is used to map all nodes from their initial coordinate space to a high-dimensional manifold space of a preset dimension, which is set to a value smaller than the initial coordinate dimension. The goal of the algorithm is to find a set of target coordinates such that all nodes with high connection weights (Weight) are aligned. uv To minimize the Euclidean distance between node pairs in the target space, we can achieve this by minimizing a loss function defined as the sum of the high-dimensional connection weights on all edges multiplied by the squared Euclidean distance between the node pairs in the target space.

[0073] First, a Laplacian matrix is ​​constructed based on the high-dimensional connection weights. The dimension of this matrix is ​​equal to the total number of functional semantic nodes. Then, the generalized eigenvalue problem of this Laplacian matrix is ​​solved to obtain the eigenvalues ​​and corresponding eigenvectors. The eigenvectors corresponding to the smallest non-zero eigenvalues ​​are selected, and the components of each node on different eigenvectors are combined to form the coordinate vector of the node in the target low-dimensional manifold space.

[0074] Step 134: In the high-dimensional manifold space, based on the high-dimensional coordinate initialization parameters of the functional semantic nodes and the high-dimensional connection weight parameters of the logical coupling relationship edges, the position optimization and adjustment processing of the functional semantic nodes is performed to generate the manifold embedding coordinates of each functional semantic node in the high-dimensional manifold space.

[0075] Using the node coordinates obtained in step 133 through the manifold dimensionality reduction and preservation algorithm as initial values, a local optimization adjustment process is performed. This process uses gradient descent to iteratively update the coordinates of each node in order to further minimize a composite loss function. This composite loss function consists of two parts: the first part is the same as the loss function in step 133, namely the weighted sum of squared distances, which is used to preserve the global structure; the second part is the mean squared error between the node and its initial coordinates, which is used to prevent the node coordinates from drifting too much and losing semantic information.

[0076] In each iteration, the gradient of the composite loss function at the current coordinates of each node is calculated, and then the node's coordinates are updated at a preset learning rate in the opposite direction of gradient descent. This iterative process is repeated until the value of the composite loss function converges to a stable minimum or the preset maximum number of iterations is reached. The final coordinates of each node obtained after the iteration are the manifold embedding coordinates of that functional semantic node in the high-dimensional manifold space.

[0077] Step 135: Based on the manifold embedding coordinates and the high-dimensional connection weight parameters, construct the local neighborhood connection relationship of each functional semantic node in the high-dimensional manifold space. The local neighborhood connection relationship reflects the geometric proximity of the process dependency strength between functional semantic nodes in the high-dimensional manifold space.

[0078] After obtaining the manifold embedding coordinates of all functional semantic nodes, for each node, the following operations are performed. First, calculate the Euclidean distance between this node and all other nodes. Then, construct a new metric called the fusion distance for each node. u and Node v The fusion distance is equal to the Euclidean distance between the two multiplied by a coefficient, which is 2 minus the high-dimensional connection weight of the logical coupling edge between them. uv .

[0079] If node u and Node v If there is no logical coupling between the edges, then Weight uv The value is 0, therefore the fusion distance is reduced based on the Euclidean distance for nodes that already have strong connections. Next, for node Node... u Select a specified number of nodes with the smallest fusion distance as its local neighbor nodes. Finally, in Node u Construct an undirected local neighborhood connection edge between each of its neighboring nodes, and assign a weight to the edge that is equal to 1 divided by 1 and the sum of the fusion distance.

[0080] Step 136: Based on the manifold embedding coordinates and local neighborhood connectivity of all functional semantic nodes, generate the global action semantic manifold structure corresponding to the complete forming cycle. The global action semantic manifold structure is a distribution configuration of action semantic nodes with topology preservation properties in a high-dimensional manifold space.

[0081] The global action semantic manifold structure is a weighted graph whose vertex set is identical to that of the process semantic topology graph. Its edge set comprises two subsets: the first subset consists of all directed logical coupling edges from the original process semantic topology graph, which are preserved but treated as undirected edges for geometric analysis; the second subset consists of all undirected local neighborhood connections constructed in step 135. Each vertex is appended with its final manifold embedding coordinates in the manifold space. This structure simultaneously encodes the deterministic logical dependencies of process execution and the implicit proximity relationships inferred based on semantic similarity.

[0082] Step 137: Calculate the homology group features of the global action semantic manifold structure in the homology group dimension. The homology group features are used to describe the number of connected voids and the distribution of ring structures in the high-dimensional manifold space of the process semantic topology graph.

[0083] Step 1371: Based on the manifold embedding coordinates of each functional semantic node in the global action semantic manifold structure, construct the simple composite form representation of the global action semantic manifold structure in the high-dimensional manifold space. The simple composite form representation is composed of simple form units of multiple dimensions connected by shared surfaces.

[0084] Extract all functional semantic nodes and their manifold embedding coordinates from the global action semantic manifold structure, as well as all edges, including edges with original logical coupling relationships and edges connecting local neighborhoods. The process of constructing the simplicial complex is as follows: treat all functional semantic nodes as 0-simplexes, each of which is uniquely identified by its node identifier.

[0085] For each connected node Node u and Node v An edge, whether directed or undirected, is in Node. u and Node v Construct a 1-simulacra between the nodes, which is represented by unordered pairs of two 0-simulacra, for example, {Node u Node v For any three functional semantic nodes (Node), ... a Node b Node c Check if there are three edges connecting Node. a With Node b Node b With Node c Node c With Node a If all three edges exist, then a 2-simplex is constructed between these three nodes. This 2-simplex is represented by unordered triplets of three 0-simplexes, for example, {Node aNode b Node c The 2-simplex is a solid triangular region. In this way, all simplex units of different dimensions are interconnected by sharing low-dimensional surfaces, forming a complete simplex complex representation. K .

[0086] Step 1372: Perform boundary operator operations on the simplex units of different dimensions in the simplex complex representation to generate the boundary relation matrix corresponding to each simplex unit. The boundary relation matrix is ​​used to describe the algebraic boundary association between the high-dimensional simplex unit and its constituent low-dimensional simplex units.

[0087] For simple complex K Define a boundary operator. For every 1-simplex, its boundary is the formal algebraic sum of its two endpoints, the 0-simplexes. For example, the boundary operator acts on the 1-simplex {Node...} u Node v The result of} is equal to Node v Subtract Node u For every 2-simplex, its boundary is the formal algebraic sum of its three edges, the sign of which is determined by the parity of the order of its vertices. For example, the boundary operator acts on the 2-simplex {Node...} a Node b Node c The result is equal to the edge {Node} b Node c} Subtract edge {Node a Node c} Add edge {Node a Node b Based on the boundary operations described above, a boundary relation matrix is ​​constructed. A matrix B1 of dimension N1 × N0 is constructed, where N1 is the number of 1-simplexes and N0 is the number of 0-simplexes. Each row of matrix B1 corresponds to a 1-simplex, and each column corresponds to a 0-simplex. The column positions corresponding to the two 0-simplexes related to the boundary in that row are filled with the values ​​1 and -1, respectively. A matrix B2 of dimension N2 × N1 is constructed, where N2 is the number of 2-simplexes and N1 is the number of 1-simplexes. Each row of matrix B2 corresponds to a 2-simplex, and each column corresponds to a 1-simplex. The column positions corresponding to the three 1-simplexes related to the boundary of that 2-simplex are filled with the values ​​1, -1, or 1, respectively, with the specific signs determined by the boundary calculations.

[0088] Step 1373: Based on the boundary relation matrices of each dimension, construct homology groups of different dimensions to calculate the kernel space and image space, and generate homology groups of each dimension by calculating the quotient space of the kernel space and the image space. The homology groups are algebraic invariants describing the number of connected holes and the distribution of ring structures in the global action semantic manifold structure.

[0089] For a 0-dimensional homology group, compute the null space of matrix B1. Vectors in the null space represent all combinations of 1-simplexes satisfying the boundary condition of zero, i.e., closed loops. Simultaneously compute the 0-dimensional boundary group, but since there is no negative 1-dimensional simplex, the 0-dimensional boundary group is considered as the null space. The 0-dimensional homology group H0 is equal to the kernel space of matrix B1.

[0090] For a 1-dimensional homology group, first calculate the image space of matrix B2, which represents all combinations of 1-simplexes that serve as the boundaries of one of the 2-simplexes. Then calculate the kernel space of matrix B1, which represents all closed 1-simplex loops. The 1-dimensional homology group H1 is equal to the quotient space obtained by dividing the kernel space of matrix B1 by the image space of matrix B2. For a 2-dimensional homology group, calculate the kernel space of matrix B2, which represents all combinations of 2-simplexes with zero boundaries, i.e., closed surfaces. Simultaneously calculate the 3-dimensional boundary group, which is considered a null space since no 3-simplex exists. The 2-dimensional homology group H2 is equal to the kernel space of matrix B2.

[0091] Step 1374: Extract the rank of the zero-dimensional homology group as the feature of the number of connected components of the global action semantic manifold structure; extract the rank of the one-dimensional homology group as the feature of the number of one-dimensional ring structures of the global action semantic manifold structure; and extract the rank of the two-dimensional homology group as the feature of the number of two-dimensional holes of the global action semantic manifold structure.

[0092] Calculate the rank of the 0-dimensional homology group H0. H0 The rank is equal to the number of generators of H0, i.e., the simple complex. K The total number of connected components is denoted as the connected component count feature Feat0. Calculate the rank of the 1-dimensional homology group H1. H1 The rank is equal to the number of generators of H1, that is, the number of independent loops in the simplex complex that cannot be filled by a 2-simplex, denoted as the one-dimensional loop number feature Feat1. Calculate the rank of the 2-dimensional homology group H2. H2 The rank is equal to the number of generators of H2, that is, the number of independent voids enclosed by 2-simplexes in the simple complex, denoted as the two-dimensional void number feature Feat2.

[0093] Step 1375: The number of connected components, the number of one-dimensional ring structures, and the number of two-dimensional voids are combined into the homology group features of the global action semantic manifold structure. The homology group features constitute the topological invariant descriptor of the mold action collaborative logic within the complete molding cycle.

[0094] The three feature values ​​extracted in step 1374 are concatenated in a fixed order to form a three-dimensional feature vector F. current =[Feat0, Feat1, Feat2], this feature vector is the homology group feature of the global action semantic manifold structure. This feature has topological invariance, that is, it remains unchanged under continuous deformation, and can characterize the essential topological structure of the collaborative logic between the various processes of the mold in the complete molding cycle.

[0095] It is understandable that the constrained Laplacian eigenmap algorithm used in step 133 has its core module as constructing the Laplacian matrix and solving the generalized eigenvalue problem. This algorithm does not involve trainable neural network parameters and belongs to the parameter-free dimensionality reduction algorithm, thus requiring no training process. In application, based on the number of nodes and edge weight matrix of the current process semantic topology graph, a Laplacian matrix L with dimension equal to the total number of nodes is directly constructed, where L = degree matrix D - adjacency weight matrix W. Then, the generalized eigenvalue problem is solved, and the eigenvectors corresponding to the smallest non-zero eigenvalues ​​are selected as the embedding coordinates.

[0096] Step 134 employs gradient descent for position optimization. This optimization process also does not involve a pre-trained model; instead, it iterative updates are performed on the node coordinates of the current batch. The composite loss function is defined as a weighted sum of two parts: the first part is the loss function value from step 133, and the second part is the mean square error between the current node coordinates and the output coordinates from step 133. The weight coefficients are optimized within the range of 0 to 1 through grid search. The optimizer uses standard stochastic gradient descent, with a learning rate of 0.01, a momentum parameter of 0.9, a maximum number of iterations of 1000, and a convergence condition set to the absolute value of the loss function change being less than a preset threshold for 20 consecutive iterations.

[0097] The homology group calculation process in steps 1372 to 1374 uses a standard algorithm in the field of algebraic topology, which is the construction of the boundary matrix and the calculation of the Smith canonical form. When applied, this algorithm directly constructs an integer coefficient boundary matrix based on the input simple complex form representation, and then calculates the Smith canonical form through row and column transformations of the integer matrix, from which the rank of each dimension of the homology group is extracted.

[0098] Step 140: Obtain the baseline action semantic manifold structure pre-constructed within the historical non-deviation molding cycle of the target injection mold, extract the baseline homology group features of the baseline action semantic manifold structure, and perform homology group perturbation analysis based on the homology group features of the global action semantic manifold structure and the baseline homology group features to generate global configuration deviation detection results.

[0099] Step 141: Obtain the reference action semantic manifold structure pre-constructed within the historical undevised molding cycle of the target injection mold. The reference action semantic manifold structure is pre-constructed by the historical action state information sequence corresponding to the historical undevised molding cycle according to the process semantic deconstruction processing and high-dimensional manifold space mapping processing.

[0100] The baseline data corresponding to the target injection mold is retrieved from the historical database. This baseline data is a representative global action semantic manifold structure obtained by independently executing all processing steps 110 to 136 on a large number of historical action state information sequences without deviation from the molding cycle.

[0101] The specific construction method is as follows: First, calculate the manifold embedding coordinates of each functional semantic node in the global action semantic manifold structure corresponding to each historical unbiased period. Then, take the arithmetic mean of the coordinates of each node in all periods as the reference manifold embedding coordinates of the node in the reference structure. Next, based on the above reference coordinates, reconstruct the local neighborhood connection relationship according to the method in step 135. Finally, the reference action semantic manifold structure Manifold is formed. bench This structure represents the standard motion cooperative geometry of the target injection mold under normal, unbiased conditions.

[0102] Step 142: Extract the baseline manifold embedding coordinates and baseline local neighborhood connectivity of each baseline functional semantic node in the baseline action semantic manifold structure. Based on the baseline manifold embedding coordinates and baseline local neighborhood connectivity, construct the baseline simple complex representation of the baseline action semantic manifold structure in the high-dimensional manifold space.

[0103] From the baseline action semantic manifold bench Read the baseline manifold embedding coordinates of each baseline functional semantic node. Read the baseline local neighborhood connectivity, which is the set of all undirected edges in the baseline structure. Construct the baseline simplicial complex in exactly the same way as in step 1371. bench All reference nodes are treated as 0-simplexes, and all reference local neighborhood connecting edges are treated as 1-simplexes. For any three reference nodes, if there are reference local neighborhood connecting edges between each pair of them, then a 2-simplex is filled between these three nodes.

[0104] Step 143: Perform homology group calculation on the reference simplex complex representation to generate a zero-dimensional reference homology group, a one-dimensional reference homology group, and a two-dimensional reference homology group. Extract the rank of the zero-dimensional reference homology group as the feature of the number of reference connected components, extract the rank of the one-dimensional reference homology group as the feature of the number of reference one-dimensional ring structures, extract the rank of the two-dimensional reference homology group as the feature of the number of reference two-dimensional holes, and combine the feature of the number of reference connected components, the feature of the number of reference one-dimensional ring structures, and the feature of the number of reference two-dimensional holes into the feature of the reference homology group.

[0105] The baseline simple complex constructed in step 142 bench Perform the same homology group calculation process as steps 1372 to 1374. First, calculate the homology group for the Complex. bench Boundary operator operations are performed on the 1-simplex and 2-simplex to construct the corresponding boundary relation matrix B1. bench and B2 bench Then, based on the above boundary matrix, calculate the 0-dimensional, 1-dimensional, and 2-dimensional homology groups.

[0106] Extracting the rank of the baseline 0-dimensional homology group as the baseline connected component number feature. 0bench Extract the rank of the benchmark 1D homology group as the numerical feature of the benchmark 1D ring structure. 1bench Extract the rank of the baseline 2D homology group as the baseline 2D hole number feature. 2bench These three baseline eigenvalues ​​are concatenated in sequence to form a three-dimensional baseline homology group eigenvector F. bench =[Feat 0bench Feat 1bench Feat 2bench ].

[0107] Step 144: Based on the homology group features of the global action semantic manifold structure and the baseline homology group features, homology group perturbation analysis is performed to generate global configuration deviation detection results. The global configuration deviation detection results include the survival interval distribution offset of the cooperative relationship between functional semantic nodes in the process semantic topology graph and the ring structure collapse degree parameter.

[0108] Step 1441: Compare the number of connected components in the homology group features of the global action semantic manifold structure with the number of connected components in the baseline homology group features to generate the change in the number of connected components.

[0109] The current homology group feature vector F obtained in step 1375 current Extracting the number of connected components feature Feat 0current The reference homology group eigenvector F obtained from step 143bench Extracting the baseline connected component number feature Feat 0bench Calculate the change in the number of connected components, Delta0 = |Feat 0current -Feat 0bench |

[0110] Step 1442: Compare the one-dimensional loop structure number feature in the homology group feature of the global action semantic manifold structure with the benchmark one-dimensional loop structure number feature in the benchmark homology group feature to generate the change in the number of one-dimensional loop structures.

[0111] From F current Extracting the number of one-dimensional ring structures (Feat) 1current From F bench Extracting the baseline one-dimensional ring structure number feature Feat 1bench Calculate the change in the number of one-dimensional ring structures, Delta1 = Feat 1current -Feat 1bench .

[0112] Step 1443: Compare the two-dimensional hole number feature in the homology group feature of the global action semantic manifold structure with the baseline two-dimensional hole number feature in the baseline homology group feature to generate the change in the number of two-dimensional holes.

[0113] From F current Extracting the two-dimensional hole number feature. 2current From F bench Extracting baseline two-dimensional hole number features Feat 2bench Calculate the change in the number of two-dimensional voids, Delta2 = Feat 2current -Feat 2bench .

[0114] Step 1444: Based on the changes in the number of connected components, the changes in the number of one-dimensional ring structures, and the changes in the number of two-dimensional holes, determine the homology group perturbation type and the homology group perturbation magnitude of the global action semantic manifold structure relative to the reference action semantic manifold structure.

[0115] The type of homology group perturbation is determined using a decision tree logic. First, it checks if the change in the number of connected components, Delta0, is greater than 0. If it is, the perturbation type is classified as topological breakage. If Delta0 equals 0, it checks if the change in the number of one-dimensional loops, Delta1, is less than 0. If it is, the perturbation type is classified as loop collapse. If Delta1 equals 0, it checks if the change in the number of two-dimensional holes, Delta2, is greater than 0. If it is, the perturbation type is classified as hole growth. If Delta2 equals 0, it checks if Delta1 is greater than 0. If it is, the perturbation type is classified as loop growth. If Delta1 equals 0, it checks if Delta2 is less than 0. If it is, the perturbation type is classified as hole collapse. If all changes are 0, the perturbation type is classified as no perturbation. The amplitude of the homology group perturbation, Amplitude, is the square root of the sum of the squares of Delta0, Delta1, and Delta2.

[0116] Step 1445: Based on the homology group perturbation type, identify the survival interval distribution offset of the cooperative relationship between functional semantic nodes in the process semantic topology graph. The survival interval distribution offset is used to characterize the degree of change in the interval length of the logical coupling relationship edge between functional semantic nodes that persists in the high-dimensional manifold space.

[0117] When the disturbance type is loop collapse or loop proliferation, the survival interval distribution offset is calculated. For each logical coupling edge in the process semantic topology graph... uv The length of its survival interval is defined as the continuous time range in which the Euclidean distance between the two nodes of the edge in the global action semantic manifold structure is less than a preset distance threshold.

[0118] In the baseline action semantic manifold bench In the process, for each corresponding edge, calculate its baseline survival interval length L. benchuv Within the current global action semantic manifold structure, calculate the length L of its current survival interval. currentuv Then the survival interval distribution offset Shift uv =(L currentuv -L benchuv ) / L benchuv Organize the offsets of the survival intervals of all logically coupled edges into an offset vector Shift according to the edge index order. vector .

[0119] Step 1446: Calculate the loop structure collapse degree parameter in the process semantic topology graph based on the homology group perturbation amplitude. The loop structure collapse degree parameter is used to characterize the degree of integrity degradation of the closed dependency loop formed by multiple functional semantic nodes in the global configuration.

[0120] When the disturbance type is loop collapse, calculate the parameters of the degree of collapse of the loop structure. First, identify the reference homology group H1. bench It exists in the current homology group H1 but in the current homology group H1 current The disappearing 1D cycle. For each disappearing cycle... K Obtain the sequence of functional semantic nodes and the sequence of logical coupling edges corresponding to each edge in the process semantic topology graph. uv Calculate the edge durability, which is equal to the current survival interval length L. currentuv Divide by the baseline survival interval length L benchuv Then, the overall durability of the cycle is calculated. The overall durability is equal to the arithmetic mean of the durability of all edges in the cycle. The degree of collapse of the cycle is then determined. K It equals 1 minus its overall durability. For all disappearing loops, calculate the weighted average of their collapse degree, with the weight being the average process dependency strength of the logical coupling edges contained in each disappearing loop. The final result is the loop structure collapse degree parameter, Collapse. para .

[0121] Step 1447: Combine the survival interval distribution offset and the ring structure collapse degree parameter into a global configuration deviation detection result.

[0122] Construct a data structure as the final global configuration deviation detection result. global The data structure contains the following fields: the perturbation type field stores the type of homology group perturbation determined in step 1444. perturb The perturbation amplitude field stores the homology group perturbation amplitude Amplitude calculated in step 1444, and the survival interval offset field stores the survival interval distribution offset vector Shift calculated in step 1445. vector The Collapse Degree field stores the Collapse Degree parameter of the ring structure calculated in step 1446. para .

[0123] In step 1444, the decision tree logic is used to determine the type of homology group perturbation. This decision tree consists of a set of hard-coded if-then rules, all with a fixed threshold of 0, requiring no training. The preset distance threshold for calculating the survival interval length in step 1445 is determined by taking the 20th percentile of the statistical distribution of the Euclidean distance between all node pairs in the baseline action semantic manifold as the threshold. In step 1446, the weighted average is calculated using the average process dependency strength of the logical coupling edges contained in each vanishing cycle, which has been pre-calculated in step 125.

[0124] Step 150: Based on the global configuration deviation detection results, determine the deviation position and deviation propagation path of the target injection mold's action timing within the complete molding cycle, and generate a mold action timing correction command based on the deviation position and deviation propagation path.

[0125] Step 151: Based on the global configuration deviation detection results, determine the deviation position and deviation propagation path of the target injection mold in the action sequence of the complete molding cycle.

[0126] Step 1511: Analyze the survival interval distribution offset in the global configuration deviation detection result, and extract the target logical coupling relationship edge in the process semantic topology graph where the survival interval distribution offset exceeds a preset offset threshold.

[0127] Set a survival interval offset threshold Thres shift Result of global configuration deviation detection global Extract the survival interval offset vector Shift vector Traverse each logical coupling edge in the process semantic topology graph. uv Query its position in Shift vector The corresponding offset value Shift uv If Shift uv The absolute value is greater than the preset offset threshold Thres shift Then mark the edge as the target logical coupling edge TargetEdge. uv The aforementioned target edges represent locations where significant anomalies occur in the temporal synchronization of action sequences.

[0128] In step 1511, a preset offset threshold Thres is set. shift The method for determining the threshold is as follows: based on the statistical distribution offset of the survival interval of all logically coupled edges in the historical normal production cycle, the 95th percentile of the distribution is taken as the threshold.

[0129] Step 1512: Determine the time position interval of the action timing deviation in the complete forming cycle based on the composite timing interval identifier of the two functional semantic nodes connected by the target logical coupling relationship edge.

[0130] For each target logical coupling relationship edge TargetEdge uv Get the preceding functional semantic node (Node) it is connected to. u Composite time interval identifier Interval u equal to the starting index u End Index u and post-functional semantic nodes Nodev Composite time interval identifier Interval v equal to the starting index v End Index v The time interval T of the action timing deviation deviation Determined to be from the end index u To the starting index v The half-open, half-closed interval, i.e., greater than the ending index. u And less than or equal to the starting index v The time period, which corresponds to Node u After the process is completed, the node v The transition phase between the start of a process is the specific time window in which timing deviations occur.

[0131] Step 1513: Analyze the ring structure collapse degree parameter in the global configuration deviation detection result to determine the closed dependency loop in which the ring structure collapse occurs. The closed dependency loop is composed of multiple functional semantic nodes and logical coupling relationship edges connected end to end.

[0132] Results of global configuration deviation detection global Extracting the Collapse parameter of the ring structure para If this parameter is greater than 0, it indicates the presence of loop collapse. By tracing back the homology group perturbation analysis process in step 1446, a list of Cycles marked as vanished 1D cycles is obtained. list For Cycle list Each disappearing cycle in K Obtain the functional semantic node sequence NodeSeq corresponding to it in the process semantic topology graph. K EdgeSeq, a sequence of logical coupling relationships K Each disappearing loop is a closed dependent loop in which the loop structure collapses.

[0133] Step 1514: Extract the composite action semantic tags and composite timing interval identifiers of each functional semantic node in the closed dependency loop, generate the process transmission path corresponding to the closed dependency loop, and determine the deviation starting functional semantic node of the action timing deviation in the process semantic topology graph based on the target logical coupling relationship edge and the process transmission path of the closed dependency loop.

[0134] For each collapsed loop Cycle identified in step 1513 K Extract its node sequence NodeSeq KEach functional semantic node in the process contains a composite action semantic label and a composite temporal interval identifier. These nodes are then arranged in ascending order according to the starting index in the composite temporal interval identifier, forming a process transfer path. K Then use the path. K All logical coupling edges on the target logical coupling edge set TargetEdge determined in step 1511 set Compare them.

[0135] In Path K The process proceeds by examining each node sequentially according to the path, finding the first node that satisfies the following conditions: either the node acts as a preceding node pointing to a target edge, or the node acts as a succeeding node pointed to by a target edge. This first found node is then designated as the deviated functional semantic node (Node) of the collapsed loop. start .

[0136] Step 1515: Starting from the deviated initial functional semantic node, trace the propagation sequence of the functional semantic node affected by the action timing deviation along the directed connection direction of the logical coupling relationship edge in the process semantic topology graph.

[0137] The Node that deviates from the initial functional semantic node determined in step 1514 start Starting from the original process semantic topology graph G, a directed breadth-first search is performed. During the search, traversal is only performed along the directed directions of logically coupled edges; that is, starting from the current node, only the successor nodes pointed to by all its outgoing edges are visited. During the traversal, a queue is maintained to record nodes to be visited and a set is maintained to record visited nodes. Node... start Add to the queue. When the queue is not empty, remove the head node CurrNode, traverse all successor nodes NextNode of CurrNode, and if NextNode has not been visited, add it to the queue and mark it as visited, while recording the successor node. start The propagation relationship to NextNode. The search terminates in this direction when it encounters another node in a collapsed loop and that node has been marked as affected, or when the search reaches an end node in the process semantic topology graph that has no outgoing edges.

[0138] Ultimately, all from Node start Nodes reachable via directed paths are arranged in chronological order of their first visit, forming the functional semantic node propagation sequence NodeSeq, which reflects the impact of action timing deviations. propagation .

[0139] Among them, the directed breadth-first search algorithm in step 1515 is a standard algorithm in graph theory. It does not involve model training and is directly used for queue traversal based on the adjacency list structure of the process semantic topology graph.

[0140] Step 1516: Combine the deviation from the initial functional semantic node and the propagation sequence of the functional semantic node into a deviation propagation path, and merge the time position interval with the deviation propagation path into an action timing deviation position and deviation propagation path that includes the time position interval and the deviation propagation path.

[0141] Construct a composite data structure Deviation info To describe the final deviation information, this structure contains a deviation start node field, StartNode, which stores the deviation start functional semantic node. start The structure also includes an identifier and a deviation propagation sequence field, PropagationSeq, used to store the functional semantic node propagation sequence NodeSeq generated in step 1515. propagation The structure also includes a TimeWindow field, which stores the time position interval T determined in step 1512. deviation Deviation info The deviation position and deviation propagation path of the final determined target injection mold during the complete molding cycle.

[0142] Step 152: Generate mold action timing correction instructions based on the deviation position and deviation propagation path.

[0143] Step 1521: Extract the time position interval in the deviation position, determine the set of deviation functional semantic nodes corresponding to the deviation position, extract the propagation sequence of functional semantic nodes in the deviation propagation path, and determine the set of affected functional semantic nodes affected by the action timing deviation.

[0144] From the deviation position and deviation propagation path determined in step 1516, Deviation info Extract the TimeWindow field from the deviation time window; the time interval corresponding to this field represents the time range in which the deviation occurred. Extract the StartNode field from the same data structure and merge it with all nodes in the PropagationSeq field to form a Set of deviation functional semantic nodes. deviation Extract all propagation nodes from PropagationSeq to form a set of affected functional semantic nodes (Set). affected .

[0145] Step 1522: Based on the set of deviated functional semantic nodes and the set of affected functional semantic nodes, extract the original composite action semantic labels and original composite time interval identifiers corresponding to the deviated functional semantic nodes and the affected functional semantic nodes from the process semantic topology graph.

[0146] Traverse the set of deviated functional semantic nodes deviation Each node in w Query the attributes of the node from the process semantic topology graph G, and extract its original composite action semantic label. origw and the original composite time interval identifier Interval origw , in the form of starting index origw End Index origw .

[0147] Similarly, traverse the Set of affected functional semantic nodes. affected Each node in x Extract its original compound action semantic label. origx and the original composite time interval identifier Interval origx .

[0148] Step 1523: Based on the reference manifold embedding coordinates of the corresponding reference function semantic node in the reference action semantic manifold structure, determine the standard composite temporal interval identifier of the deviated function semantic node and the affected function semantic node in the undeviated state.

[0149] For the set of nodes that deviate from functional semantics deviation Each node in w In the baseline action semantic manifold structure Manifold bench Find the corresponding baseline functional semantic node in the middle. wbench Since the node set of the process semantic topology graph and the node set of the baseline structure have the same semantic definition, matching can be performed using composite action semantic tags. From Node wbench Extract its baseline composite time interval identifier Interval stdw , in the form of starting index stdw End Index stdw .

[0150] Similarly, for the set of affected functional semantic nodes Set affected Each node in x At Manifold bench Find the corresponding node in xbench Extract its standard composite time interval identifier Interval stdx .

[0151] Step 1524: Perform timing deviation calculation on the original composite timing interval identifier and the standard composite timing interval identifier to generate the timing correction offset of each deviated functional semantic node and the affected functional semantic node.

[0152] For each functional semantic node that needs correction y , where Node y Belongs to Set deviation and Set affected The union of the original composite time intervals is used to obtain their original identifiers. origy equal to the starting index origy End Index origy and its standard composite time interval identifier Interval stdy equal to the starting index stdy End Index stdy Timing correction offset y It is a two-dimensional vector whose first component Delta starty equal to the starting index stdy Subtract the starting index origy Its second component Delta endy equal to the end index stdy Subtract the end index origy Delta starty The positive or negative sign of Delta indicates whether the node needs to start execution earlier or later. endy The positive or negative sign indicates whether the node needs to end execution earlier or later.

[0153] Step 1525: Generate a mold action timing correction instruction based on the timing correction offset and the original composite action semantic tag corresponding to the deviation functional semantic node set. The mold action timing correction instruction includes the functional semantic node identifier to be corrected, the target timing interval identifier after correction, and the action execution order adjustment instruction.

[0154] All functional semantic nodes that need correction y The node identifier and the corresponding original compound action semantic label. origy And the calculated timing correction offset y Combined into a correction record list list For each corrected record, calculate the corrected target time interval identifier, Interval. targety : Starting index targety equal to the starting index origy Add Delta startyEnd of index targety equal to the end index origy Add Delta endy Generate a mold action timing correction instruction. The data structure of this instruction includes an instruction type field with a value of timing offset correction, and a correction record list field storing the correction. list Each record contains a node identifier, a target start index, and a target end index.

[0155] In addition, based on the timing offsets in the correction record list, an action execution order adjustment instruction is generated. This instruction is a Boolean flag; if any Delta exists... starty A negative value indicates that the corresponding action needs to be performed in advance; if any Delta exists... starty A positive value indicates that the corresponding action needs to be delayed.

[0156] Step 1526: Send the mold action timing correction command to the control system of the target injection mold to trigger the action timing adjustment operation of the target injection mold in the subsequent molding cycle.

[0157] The mold action timing correction command generated in step 1525 is encapsulated into a data packet conforming to the communication protocol of the target injection mold control system via the industrial Ethernet protocol and sent to the command receiving port of the control system. The control system parses the command, extracts the correction record list, and in each subsequent molding cycle, when the corresponding functional semantic node is executed, the trigger time and duration of the action are adjusted according to the corrected target timing interval identifier, thereby achieving closed-loop correction of the action timing.

[0158] Based on step 150, the method further includes: Step 210: Obtain the sequence of subsequent action status information collected in multiple subsequent complete molding cycles after the target injection mold implements the mold action timing correction command.

[0159] After the mold action timing correction command generated in step 1526 is sent to the control system of the target injection mold and successfully executed, the same data acquisition interface established in step 110 is used to continuously acquire the action status information sequence of the target injection mold in each of the subsequent multiple complete molding cycles. The sequence acquisition method, sampling time interval, and format of the action status information unit for each subsequent cycle are exactly the same as in step 110.

[0160] Step 220: For each subsequent complete forming cycle, the subsequent action state information sequence is processed according to the semantic association and reorganization to generate a subsequent process semantic topology graph, and the subsequent process semantic topology graph is mapped to the high-dimensional manifold space to construct the subsequent global action semantic manifold structure.

[0161] For each subsequent complete forming cycle's action state information sequence obtained in step 210, all processing logic described in steps 120 to 136 is executed independently and completely. First, the semantic association reorganization in steps 120 to 127 is performed on the sequence of each subsequent cycle to generate the subsequent process semantic topology graph corresponding to that cycle. Then, the high-dimensional manifold space mapping and embedded coordinate optimization in steps 130 to 136 are performed on each subsequent process semantic topology graph to construct the subsequent global action semantic manifold structure corresponding to that cycle.

[0162] Step 230: Perform homology group calculation on each subsequent global action semantic manifold structure to extract subsequent homology group features that include the number of subsequent connected components, the number of subsequent one-dimensional ring structures, and the number of subsequent two-dimensional holes.

[0163] For each subsequent global action semantic manifold structure constructed in step 220, the homology group computation processing described in steps 1371 to 1375 is performed independently and completely. The rank of the 0-dimensional homology group of each subsequent structure is extracted as the feature of the number of subsequent connected components. 0sub Extract the rank of its one-dimensional homology group as the subsequent one-dimensional ring structure number feature. 1sub Extract the rank of its 2D homology group as the subsequent 2D hole number feature. 2sub These three eigenvalues ​​are concatenated in sequence into a three-dimensional vector, which serves as the feature F of the subsequent homology group corresponding to that subsequent period. sub =[Feat 0sub Feat 1sub Feat 2sub ].

[0164] Step 240: Construct a homology group feature evolution sequence according to the cycle order of all subsequent complete molding cycles, and perform cross-cycle homology group difference distribution analysis on the homology group feature evolution sequence and the reference homology group feature. By identifying the convergence or divergence trends of the subsequent connected component number feature, the subsequent one-dimensional ring structure number feature, and the subsequent two-dimensional void number feature in continuous cycles, generate manifold configuration recovery trend description information of the mold action timing correction command.

[0165] The subsequent homology group features F obtained in step 230 for all subsequent periods are... sub Arranged chronologically according to the formation cycle, a homology group feature evolution sequence Seq is formed. F =[F sub1 F sub2 F subT], where T is the total number of subsequent periods. For each subsequent homology group feature F in the sequence subT Calculate its characteristic F with the benchmark homology group. bench The difference vector D between them t D t Each component equals F subT Subtract F from the corresponding component bench The absolute values ​​of the corresponding components. Analyze the difference vector sequence [D1, D2, ..., D...]. T The trend of change within continuous periods. For each dimension d, the first-order difference of the difference components between adjacent periods is calculated. If the first-order differences of all dimensions are negative in multiple consecutive periods, the dimension is determined to show a convergent trend. If the first-order difference of any dimension is positive and continuously increasing, or shows irregular oscillations of alternating positive and negative values, the dimension is determined to show a divergent trend. The results of all dimensions are combined to generate manifold configuration recovery trend description information: Recovery. desc This information is a text label. If all dimensions show a convergence trend, the value is that the topology is recovering; if any dimension shows a divergence trend, the value is that the topology is deviating further.

[0166] As an optional embodiment, after generating the mold action timing correction command based on the deviation position and deviation propagation path, the method further includes: Step 310: Extract the functional semantic nodes along the deviation propagation path from the process semantic topology graph to form a set of deviation nodes, and identify the original logical coupling relationship edges between each functional semantic node in the deviation node set in the process semantic topology graph.

[0167] From the deviation position and deviation propagation path determined in step 1516, Deviation info Extract the PropagationSeq field from the deviance propagation sequence. This sequence contains all functional semantic nodes traversed along the deviance propagation path. Store these nodes in a Set. devpath Then, in the original process semantic topology graph G, find all pairs of endpoints that belong to Set. devpath The directed logical coupling edges are stored in the set Edges. origdev The edges described above constitute the original cooperative relationships within the set of deviating nodes when executing the deviating path.

[0168] Step 320: Based on the reference local neighborhood connection relationship of the corresponding reference functional semantic node in the reference action semantic manifold structure, perform directed decoupling processing on the original logical coupling relationship edge, and generate redundant logical coupling relationship edges associated with functional semantic nodes in the deviation node set based on the neighborhood dependency structure of the decoupled functional semantic node.

[0169] Obtain the baseline action semantic manifold structure Manifold bench Set of all nodes that are offset from the target devpath The corresponding baseline functional semantic nodes within the nodes form a baseline node set Set. benchpath Extract the local neighborhood connections between the aforementioned baseline nodes, i.e., all connected sets. benchpath The undirected edges between two nodes are stored in the set Edges. benchneighbor .

[0170] For the set of off-nodes Set devpath Each pair of nodes in u and Node v If they have corresponding local neighborhood connection edges in the baseline structure, i.e. (Node ubench Node vbench (belongs to Edges) benchneighbor However, if there is no corresponding directed logical coupling edge in the original process semantic topology graph, or if there is an edge but it is marked as the target logical coupling edge in step 1511, then a new redundant logical coupling edge, RedunEdge, is generated based on the baseline local neighborhood connection relationship. uv The redundant edge is set to connect in a bidirectional direction to represent its flexibility and substitutability in collaboration, and the weight of the edge is set to the weight value of the corresponding local neighborhood connection edge.

[0171] Step 330: Integrate the redundant logical coupling relationship edges into the process semantic topology graph to obtain a redundant collaborative process semantic topology graph, and generate redundant collaborative execution logic for mold action timing based on the functional semantic node sequence connected by the redundant logical coupling relationship edges in the redundant collaborative process semantic topology graph.

[0172] Add all the redundant logical coupling edges (RedunEdge) generated in step 320 to the original process semantic topology graph G to form a new graph structure called the redundant collaborative process semantic topology graph G. redun In this new graph, for any two nodes deviating from the propagation path, if there is a redundant logical coupling edge between them, then it is considered that there exists an alternative action connection path that does not depend on the original deviation path. The sequence of functional semantic nodes connected by the aforementioned redundant edges is denoted as Set, representing the set of deviation nodes. devpath Each node in z Identify at least one alternative collaborative relationship that does not pass through the critical edge of the original off-path.

[0173] The aforementioned alternative coordination relationships are logically encapsulated to generate redundant coordinated execution logic Redun for mold action timing. logic This logic can be described as a set of conditional triggering rules: when the offset of the survival interval on the original logical coupling relationship edge is detected to exceed the preset alarm threshold, the corresponding redundant logical coupling relationship edge is automatically activated, and the action execution order is re-planned according to the sequence of alternative nodes connected by the redundant edge, so as to maintain the topological integrity of action coordination by calling the redundant logical coupling relationship edge when the deviation trend recurs.

[0174] The redundant collaborative logic construction in steps 310 to 330 is essentially edge matching and graph addition operations based on the baseline local neighborhood connection relationship. All rules are deterministic logic and require no training. When applying all the above non-model algorithms, the preprocessing method for the input data is to directly use the intermediate results calculated and stored in memory or database in steps 110 to 136, without additional standardization or normalization processing. The output results are directly used for calculation in subsequent steps.

[0175] See Figure 2 The overall principle of the injection mold action timing deviation detection method based on state sequence semantic encoding provided in this embodiment of the invention is as follows: First, for the operation of the target injection mold throughout the complete molding cycle, a sensor network consisting of displacement sensors, speed sensors, pressure sensors, and temperature sensors deployed on key moving parts of the mold continuously collects multi-channel sensor data at fixed sampling time intervals. The multi-dimensional sensor data collected at each sampling moment constitutes an action state information unit. All action state information units at all sampling moments are arranged in ascending order of time to form an action state information sequence. This sequence completely records the microscopic state changes of the mold throughout the entire molding cycle, from mold closing, through injection, holding pressure, plasticizing, mold opening, ejection, until preparation for the next mold closing.

[0176] Secondly, the acquired action state information sequence is used as input for semantic reorganization processing. Specifically, the action temporal position identifier and action type label of each action state information unit are extracted from the action state information sequence. Adjacent unit pairs that satisfy the action execution mechanism dependency constraint and action transmission direction constraint are identified. These adjacent unit pairs are semantically fused to generate composite action semantic labels and composite temporal interval identifiers, thus forming functional semantic nodes. Then, the process dependency strength and action connection constraints between different functional semantic nodes are extracted to generate logical coupling relationship edges connecting two functional semantic nodes. All functional semantic nodes and all logical coupling relationship edges are combined into a process semantic topology graph. This graph structure uses functional semantic nodes to represent macroscopic process transmission semantic units and logical coupling relationship edges to represent the temporal dependency and cooperation constraint relationships between process units, transforming the original time series data into a graph structure with clear process meaning.

[0177] Next, the process semantic topology graph is mapped to a high-dimensional manifold space to construct a global action semantic manifold structure. The composite action semantic label and composite temporal interval identifier of each functional semantic node are extracted to construct high-dimensional coordinate initialization parameters. Simultaneously, the connection strength and constraint type of each logical coupling edge are extracted to construct high-dimensional connection weight parameters. A manifold dimensionality reduction and preservation algorithm is used to map all nodes to the high-dimensional manifold space, and the manifold embedding coordinates of each node are generated through position optimization adjustments, thereby constructing local neighborhood connections and ultimately forming a global action semantic manifold structure. Based on this manifold structure, its homology group features are calculated, including the rank of the zero-dimensional homology group as a feature of the number of connected components, the rank of the one-dimensional homology group as a feature of the number of one-dimensional ring structures, and the rank of the two-dimensional homology group as a feature of the number of two-dimensional voids. Homology group features, as topological invariants, can describe the number of connected voids and the distribution of ring structures in the high-dimensional manifold space of the process semantic topology graph, characterizing the essential topological structure of the mold action collaborative logic within the complete molding cycle.

[0178] Then, homology group perturbation analysis is performed. The baseline action semantic manifold structure of the target injection mold, pre-constructed within a historical undevised molding cycle, is obtained, and its baseline homology group features are extracted. The homology group features of the current cycle are compared with the baseline homology group features to generate changes in the number of connected components, the number of one-dimensional ring structures, and the number of two-dimensional voids, thereby determining the homology group perturbation type and amplitude. Based on the perturbation type, the survival interval distribution offset of the collaborative relationships between functional semantic nodes is identified. The ring structure collapse degree parameter is calculated based on the perturbation amplitude. The survival interval distribution offset and the ring structure collapse degree parameter are combined into a global configuration deviation detection result, which quantifies the topological difference between the mold action collaborative logic in the current molding cycle and the historical undevised baseline.

[0179] Finally, based on the global configuration deviation detection results, the deviation position and deviation propagation path of the target injection mold's action timing within the complete molding cycle are determined. The distribution offset of the survival interval is analyzed, and the target logical coupling relationship edges with offsets exceeding a preset threshold are extracted. The time position interval of the deviation is determined based on the composite timing interval identifiers of the two functional semantic nodes connected to them. The collapse degree parameters of the ring structure are analyzed to determine the closed dependency loops where collapse occurs. The composite action semantic labels and composite timing interval identifiers of each node in the loop are extracted to generate the process transmission path. Combined with the target logical coupling relationship edges, the deviation's starting functional semantic node is determined. Starting from the deviation's starting node, the propagation sequence of functional semantic nodes affecting the deviation is traced along the directed connection direction of the logical coupling relationship edges in the process semantic topology diagram. The deviation's starting node and propagation sequence are combined into a deviation propagation path. Based on the deviation position and deviation propagation path, a mold action timing correction command is generated and sent to the target injection mold's control system to trigger action timing adjustment operations in subsequent molding cycles.

[0180] Through the above technical solution, the present invention achieves accurate identification of the timing deviation of injection mold actions at the global topology configuration level. It can determine the starting position, influence range and propagation path of the deviation, and generate targeted correction instructions, providing reliable technical support for intelligent monitoring and closed-loop control of mold action timing.

[0181] In summary, the solution provided by this invention generates a process semantic topology graph that can characterize the inter-process dependencies by semantically recombining the sequence of action state information continuously collected during the injection mold molding cycle according to the molding process logic. This process semantic topology graph is then mapped to a high-dimensional manifold space to construct a global action semantic manifold structure, and its homology group features are calculated. This achieves a topological invariant description of the mold action collaboration logic, enabling complex action collaboration relationships to be quantified into stable algebraic features. By introducing a pre-constructed baseline action semantic manifold structure and its baseline homology group features from historical undevised cycles, and performing perturbation analysis with the homology group features of the current cycle, a global configuration deviation detection result containing parameters such as the survival interval distribution offset and the degree of ring structure collapse can be generated. This accurately captures abnormal changes in the collaboration relationships between functional semantic nodes in the process semantic topology graph. Finally, based on the deviation detection results, the deviation position and deviation propagation path of the action sequence are determined, and correction instructions are generated. This achieves the identification of the starting point, influence range, and propagation direction of action sequence anomalies from the global topology configuration level, thereby realizing precise control and intelligent operation and maintenance of the injection mold.

[0182] This invention also provides a timing deviation detection server for injection mold actions: a processor; a storage device storing a computer program thereon; and a network interface for providing network communication functions; when the computer program is executed by the processor, the processor implements any of the aforementioned timing deviation detection methods for injection mold actions based on state sequence semantic encoding.

[0183] Figure 3 Also shown is one exemplary block diagram of an injection mold action timing deviation detection server 300, which has one or more processors 302, a control module (chipset) 304 coupled to at least one of the processors 302, a memory 306 coupled to the control module 304, a non-volatile memory (NVM) / storage device 308 coupled to the control module 304, one or more input / output devices 310 coupled to the control module 304, and a network interface 312 coupled to the control module 304. The memory 306 and the NVM / storage device 308 can be used to store data and / or instructions 314.

[0184] Based on the above, a readable storage medium is provided, on which a program or instructions are stored, and when the program or instructions are executed by a processor, the steps of the above method are implemented.

[0185] Furthermore, it should be noted that this embodiment of the invention also provides a computer program product, which may include a computer program that can be stored in a computer-readable storage medium. The processor of the injection mold action timing deviation detection server reads the computer program from the computer-readable storage medium, and the processor can execute the computer program, causing the injection mold action timing deviation detection server to execute the aforementioned... Figure 1 The methods described in the corresponding embodiments are already known, and therefore will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated. For technical details not disclosed in the computer program product embodiments related to this invention, please refer to the description of the method embodiments of this invention.

[0186] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

Claims

1. A method for detecting timing deviations in injection mold actions based on state sequence semantic encoding, characterized in that, include: The sequence of action status information collected sequentially over time during the complete molding cycle of the target injection mold is obtained, and the sequence of action status information contains multiple action status information units. The action state information units in the action state information sequence are semantically associated and reorganized according to the mold forming process logic to generate a process semantic topology graph containing functional semantic nodes and logical coupling relationship edges. Based on the functional semantic nodes and logical coupling relationship edges of the process semantic topology graph, the process semantic topology graph is mapped to a high-dimensional manifold space to construct the global action semantic manifold structure corresponding to the complete forming cycle, and the homology group feature of the global action semantic manifold structure in the homology group dimension is calculated. The reference action semantic manifold structure of the target injection mold is pre-constructed within the historical non-deviation molding cycle. The reference homology group features of the reference action semantic manifold structure are extracted. Based on the homology group features of the global action semantic manifold structure and the reference homology group features, homology group perturbation analysis is performed to generate global configuration deviation detection results. The global configuration deviation detection results include the survival interval distribution offset of the cooperation relationship between functional semantic nodes in the process semantic topology graph and the ring structure collapse degree parameter. Based on the global configuration deviation detection results, the deviation position and deviation propagation path of the target injection mold in the complete molding cycle are determined, and a mold action timing correction command is generated based on the deviation position and deviation propagation path.

2. The method according to claim 1, characterized in that, The functional semantic nodes are formed by fusing adjacent action state information units with direct driving and transmission relationships. The logical coupling relationship edges are used to characterize the process dependency strength and action connection constraints between different functional semantic nodes. The step of semantically associating and reorganizing the action state information units in the action state information sequence according to the mold forming process logic to generate a process semantic topology graph containing functional semantic nodes and logical coupling relationship edges includes: Extract the action timing position identifier and action type label of each action state information unit in the action state information sequence, and determine the temporal order relationship and action category attribute of each action state information unit in the complete forming cycle; Based on the time sequence relationship and action category attribute, identify adjacent action state information unit pairs in the action state information sequence that satisfy the preset mold action process constraints. The preset mold action process constraints include action execution mechanism dependency constraints and action transmission direction constraints. The adjacent action state information unit pairs that satisfy the preset mold action process constraints are semantically fused, the action type tags of the adjacent action state information unit pairs are merged into composite action semantic tags, and the action temporal position identifiers of the adjacent action state information unit pairs are combined into composite temporal interval identifiers. The functional semantic node is generated based on the composite action semantic tag and the composite time interval identifier. The functional semantic node corresponds to a complete process transmission semantic unit in the mold action process. Extract the process dependency strength and action connection constraints between different functional semantic nodes under the mold forming process logic. The process dependency strength is determined according to the pre- and post-process dependencies of the corresponding process of the functional semantic node in the forming cycle. The action connection constraints are determined according to the switching order of the action execution mechanism and the action transmission time interval of the corresponding process of the functional semantic node. The logical coupling relationship edge is generated based on the process dependency strength and action connection constraint. The logical coupling relationship edge connects two functional semantic nodes with process dependency relationship and is labeled with connection strength and constraint type. Combine all functional semantic nodes and all logical coupling relationship edges into a process semantic topology graph.

3. The method according to claim 1, characterized in that, The process semantic topology graph is mapped to a high-dimensional manifold space based on the functional semantic nodes and logical coupling edges of the process semantic topology graph, thereby constructing the global action semantic manifold structure corresponding to the complete forming cycle, including: Extract the composite action semantic label and composite temporal interval identifier of each functional semantic node in the process semantic topology diagram, and construct the high-dimensional coordinate initialization parameters of each functional semantic node. The high-dimensional coordinate initialization parameters include action semantic dimension coordinates and temporal interval dimension coordinates. Extract the connection strength and constraint type of each logical coupling relationship edge in the process semantic topology graph, and construct a high-dimensional connection weight parameter for the logical coupling relationship edge. The high-dimensional connection weight parameter is used to characterize the distance constraint relationship between functional semantic node pairs in the high-dimensional manifold space. Based on the high-dimensional coordinate initialization parameters and the high-dimensional connection weight parameters, a manifold dimensionality reduction and preservation algorithm is used to map all functional semantic nodes of the process semantic topology graph to a high-dimensional manifold space of a preset dimension. In the high-dimensional manifold space, based on the high-dimensional coordinate initialization parameters of the functional semantic nodes and the high-dimensional connection weight parameters of the logical coupling relationship edges, the position optimization and adjustment processing of the functional semantic nodes is performed to generate the manifold embedding coordinates of each functional semantic node in the high-dimensional manifold space. Based on the manifold embedding coordinates and the high-dimensional connection weight parameters, local neighborhood connection relationships of each functional semantic node in the high-dimensional manifold space are constructed. The local neighborhood connection relationships reflect the geometric proximity of the process dependency strength between functional semantic nodes in the high-dimensional manifold space. Based on the manifold embedding coordinates and local neighborhood connectivity of all functional semantic nodes, a global action semantic manifold structure corresponding to the complete forming cycle is generated. The global action semantic manifold structure is a distribution configuration of action semantic nodes with topology preservation properties in a high-dimensional manifold space.

4. The method according to claim 3, characterized in that, The homology group features are used to describe the number of connected voids and the distribution of ring structures in the high-dimensional manifold space of the process semantic topology graph. The calculation of the homology group features of the global action semantic manifold structure in the homology group dimension includes: Based on the manifold embedding coordinates of each functional semantic node in the global action semantic manifold structure, a simple composite form representation of the global action semantic manifold structure in the high-dimensional manifold space is constructed. The simple composite form representation is composed of simple form units of multiple dimensions connected by shared surfaces. Boundary operator operations are performed on the simplex units of different dimensions in the simplex complex representation to generate the boundary relation matrix corresponding to each simplex unit of each dimension. The boundary relation matrix is ​​used to describe the algebraic boundary relationship between the high-dimensional simplex unit and its constituent low-dimensional simplex units. Based on the boundary relation matrices of each dimension, homology groups of different dimensions are constructed to calculate the kernel space and image space. Homology groups of each dimension are generated by calculating the quotient space of the kernel space and the image space. The homology groups are algebraic invariants describing the number of connected holes and the distribution of ring structures in the global action semantic manifold structure. The rank of the zero-dimensional homology group is extracted as the feature of the number of connected components of the global action semantic manifold structure; the rank of the one-dimensional homology group is extracted as the feature of the number of one-dimensional ring structures of the global action semantic manifold structure; and the rank of the two-dimensional homology group is extracted as the feature of the number of two-dimensional holes of the global action semantic manifold structure. The number of connected components, the number of one-dimensional ring structures, and the number of two-dimensional voids are combined to form the homology group features of the global action semantic manifold structure. The homology group features constitute the topological invariant descriptor of the mold action collaborative logic within the complete molding cycle.

5. The method according to claim 1, characterized in that, The step of obtaining the baseline action semantic manifold structure of the target injection mold pre-constructed within a historical, undevised molding cycle, and extracting the baseline homology group features of the baseline action semantic manifold structure, includes: The reference action semantic manifold structure of the target injection mold is pre-constructed within the historical undevised molding cycle. The reference action semantic manifold structure is pre-constructed by the historical action state information sequence corresponding to the historical undevised molding cycle according to the process semantic deconstruction and high-dimensional manifold space mapping. Extract the baseline manifold embedding coordinates and baseline local neighborhood connectivity of each baseline functional semantic node in the baseline action semantic manifold structure. Based on the baseline manifold embedding coordinates and baseline local neighborhood connectivity, construct the baseline simple complex representation of the baseline action semantic manifold structure in the high-dimensional manifold space. The reference simplex complex representation is processed by homology group calculation to generate a zero-dimensional reference homology group, a one-dimensional reference homology group, and a two-dimensional reference homology group. The rank of the zero-dimensional reference homology group is extracted as the feature of the number of reference connected components. The rank of the one-dimensional reference homology group is extracted as the feature of the number of reference one-dimensional ring structures. The rank of the two-dimensional reference homology group is extracted as the feature of the number of reference two-dimensional holes. The feature of the number of reference connected components, the feature of the number of reference one-dimensional ring structures, and the feature of the number of reference two-dimensional holes are combined into the feature of the reference homology group.

6. The method according to claim 1 or 5, characterized in that, The homology group perturbation analysis based on the homology group features of the global action semantic manifold structure and the baseline homology group features generates a global configuration deviation detection result, including: The number of connected components in the homology group features of the global action semantic manifold structure is compared with the number of connected components in the benchmark homology group features to generate the change in the number of connected components. The number of one-dimensional loop structures in the homology group features of the global action semantic manifold structure is compared with the number of one-dimensional loop structures in the benchmark homology group features to generate the change in the number of one-dimensional loop structures. The two-dimensional hole number feature in the homology group feature of the global action semantic manifold structure is compared with the baseline two-dimensional hole number feature in the baseline homology group feature to generate the change in the number of two-dimensional holes. Based on the changes in the number of connected components, the changes in the number of one-dimensional ring structures, and the changes in the number of two-dimensional holes, the type and magnitude of the homology group perturbation of the global action semantic manifold structure relative to the baseline action semantic manifold structure are determined. Based on the homology group perturbation type, the survival interval distribution offset of the cooperative relationship between functional semantic nodes in the process semantic topology graph is identified. The survival interval distribution offset is used to characterize the degree of change in the interval length of the logical coupling relationship edge between functional semantic nodes that persists in the high-dimensional manifold space. Based on the amplitude of the homology group perturbation, the collapse degree parameter of the ring structure in the process semantic topology graph is calculated. The collapse degree parameter of the ring structure is used to characterize the degree of integrity degradation of the closed dependency loop formed by multiple functional semantic nodes in the global configuration. The survival interval distribution offset and the ring structure collapse degree parameter are combined to form the global configuration deviation detection result.

7. The method according to claim 1, characterized in that, The step of determining the deviation position and deviation propagation path of the target injection mold in the complete molding cycle based on the global configuration deviation detection results includes: Analyze the survival interval distribution offset in the global configuration deviation detection result, and extract the target logical coupling relationship edge in the process semantic topology graph whose survival interval distribution offset exceeds a preset offset threshold. Based on the composite temporal interval identifier of the two functional semantic nodes connected by the target logical coupling relationship edge, the time position interval of the action timing deviation in the complete forming cycle is determined. The parameters of the degree of ring structure collapse in the global configuration deviation detection results are analyzed to determine the closed dependency loops in which the ring structure collapses. The closed dependency loops are composed of multiple functional semantic nodes and logical coupling relationship edges connected end to end. Extract the composite action semantic tags and composite timing interval identifiers of each functional semantic node in the closed dependency loop, generate the process transmission path corresponding to the closed dependency loop, and determine the deviation of the action timing from the deviation starting functional semantic node in the process semantic topology graph based on the target logical coupling relationship edge and the process transmission path of the closed dependency loop. Starting from the deviated initial functional semantic node, the propagation sequence of functional semantic nodes affected by the action timing deviation is traced along the directed connection direction of the logical coupling relationship edge in the process semantic topology graph. The deviation from the initial functional semantic node and the propagation sequence of the functional semantic node are combined into a deviation propagation path, and the time position interval and the deviation propagation path are merged into an action timing deviation position and deviation propagation path that includes the time position interval and the deviation propagation path.

8. The method according to claim 1 or 7, characterized in that, The generation of mold motion timing correction instructions based on the deviation position and deviation propagation path includes: Extract the time position interval in the deviation position, determine the set of deviation functional semantic nodes corresponding to the deviation position, extract the propagation sequence of functional semantic nodes in the deviation propagation path, and determine the set of affected functional semantic nodes affected by the action timing deviation; Based on the set of deviated functional semantic nodes and the set of affected functional semantic nodes, extract the original composite action semantic labels and original composite time interval identifiers corresponding to the deviated functional semantic nodes and affected functional semantic nodes from the process semantic topology graph; Based on the reference manifold embedding coordinates of the corresponding reference functional semantic nodes in the reference action semantic manifold structure, the standard composite temporal interval identifiers of the deviated functional semantic nodes and the affected functional semantic nodes in the undeviated state are determined. The original composite time series interval identifier and the standard composite time series interval identifier are processed to calculate the time series deviation, and the time series correction offset of each deviated functional semantic node and the affected functional semantic node is generated. Based on the timing correction offset and the original composite action semantic tag corresponding to the set of deviation functional semantic nodes, a mold action timing correction instruction is generated. The mold action timing correction instruction includes the identifier of the functional semantic node to be corrected, the identifier of the target timing interval after correction, and the action execution order adjustment instruction. The mold action timing correction command is sent to the control system of the target injection mold to trigger the action timing adjustment operation of the target injection mold in the subsequent molding cycle.

9. A timing deviation detection server for injection mold actions, characterized in that, include: A processor; a storage device having a computer program stored thereon; a network interface for providing network communication functions; when the computer program is executed by the processor, the processor enables the processor to implement the injection mold action timing deviation detection method based on state sequence semantic encoding as described in any one of claims 1-8.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions, which, when executed by a processor, implement the injection mold action timing deviation detection method based on state sequence semantic encoding as described in any one of claims 1-8.

Images