Factory production intelligent supervision and management method and system

Abstract

The invention discloses an intelligent supervision and management method and system for factory production, and relates to the technical field of production management, and the method comprises the steps: constructing a time difference chain based on an operation behavior, and extracting a fluctuation feature element through a sliding window; performing frequency domain transformation on the fluctuation characteristic elements to obtain a rhythm fluctuation spectrum; calculating a dynamic time warping distance between the rhythm fluctuation spectrum and a historical rhythm reference spectrum to obtain a spectrum deviation descriptor; mapping the spectral deviation descriptors into a standard semantic process flow diagram in a nested manner to form a rhythm process nested diagram; node-level index semantic vectors are extracted from the rhythm process nested graph based on a graph embedding algorithm, multi-dimensional semantic alignment detection is executed, and alignment residual measurement is output; and aggregating the alignment residual measurement, generating a structured dislocation score list, and constructing an abnormal behavior scene graph based on a score threshold. According to the method, the rhythm fluctuation spectrum is generated and semantic nested mapping is carried out, so that the abnormity and semantic dislocation in the production supervision process are identified.

Description

Technical Field

[0001] The present invention relates to the technical field of production management, and in particular to a method and system for intelligent factory production supervision and management. Background Art

[0002] With the deepening development of digital and intelligent industrial manufacturing, factory production management systems have gradually achieved real-time monitoring and scheduling of operational processes, equipment status, and production resources. Based on operation logs, sensor data, and task scheduling information, existing manufacturing execution systems, industrial IoT platforms, and rule-based event processing engines can now track key processes and generate anomaly alerts, effectively supporting production continuity and efficiency. Simultaneously, information modeling, graph construction, and rule-based monitoring mechanisms are continuously advancing the visualization and standardization of production processes.

[0003] However, in complex real-world production scenarios, operational behaviors exhibit certain rhythmic fluctuations and flexible execution paths. Traditional methods remain challenging in capturing nonlinear behavioral rhythm patterns, quantifying operational deviations, and aligning process semantic structures. Especially when considering multidimensional factors such as temporal dynamics, behavioral rhythms, and process path semantics, establishing a unified modeling system and enabling the nested fusion and intelligent analysis of multi-scale data has become a key technical challenge in the field of production supervision and management. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a factory production intelligent supervision and management method to solve the problem of how to achieve nested fusion and intelligent analysis of multi-scale data.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for intelligent supervision and management of factory production, comprising: constructing a time difference chain based on operational behavior, extracting statistical characteristics of time intervals and behavioral trends within a sliding window, and generating a fluctuation feature primitive representing the operational rhythm; performing a frequency domain transformation on the fluctuation feature primitive, extracting frequency domain energy distribution characteristics, and generating a rhythm fluctuation spectrum;

[0008] Calculate the dynamic time warping distance between the rhythm fluctuation spectrum and the historical rhythm reference spectrum to obtain the spectrum deviation descriptor;

[0009] Assign deviation weight values, and nest and map the spectrum deviation descriptors into the standard semantic process flow diagram according to the deviation weight values ​​to form a rhythm process nested diagram;

[0010] Extract node-level indicator semantic vectors from the rhythm process nested graph based on graph embedding algorithm, perform multi-dimensional semantic alignment detection, and output alignment residual metrics;

[0011] Aggregate the alignment residual metrics to generate a structured misalignment score list, and build an abnormal behavior scenario graph based on the score threshold.

[0012] As a preferred embodiment of the factory production intelligent supervision and management method of the present invention, the spectrum deviation descriptor is used to characterize the overall difference between the rhythm fluctuation spectrum in the current window and the historical rhythm reference spectrum in terms of the main frequency position, high-frequency energy distribution, and spectrum morphology characteristics;

[0013] Through the dynamic time warping algorithm, the frequency components of the rhythm fluctuation spectrum and the historical rhythm reference spectrum are aligned segment by segment, the cumulative difference values ​​of the corresponding frequency amplitudes on the alignment path are calculated, and the deviation feature vector containing time offset compensation is generated as the spectrum deviation descriptor based on the difference distribution.

[0014] As a preferred solution of the factory production intelligent supervision and management method of the present invention, wherein: the deviation weight value is assigned, and the spectrum deviation descriptor is nested and mapped into the standard semantic process flow chart according to the deviation weight value to form a rhythm process nested chart, the specific steps are as follows:

[0015] Parse the node labels in the standard semantic process flow diagram and match the process stage identifiers associated with the spectral deviation descriptors;

[0016] The deviation feature vector is assigned to the corresponding node according to the process stage weight, and the deviation value and timestamp association relationship are recorded in the node attributes to form a rhythm process nested graph.

[0017] As a preferred solution of the factory production intelligent supervision and management method of the present invention, wherein: the time difference chain is composed of the time intervals between consecutive operation behaviors, and is marked in combination with the operation behavior category and the workstation identification;

[0018] The fluctuation feature primitive sets a fixed time window, performs sliding extraction on the time difference chain, and combines the time interval mean, time interval variance, time interval range and the differential trend between adjacent windows within the window to generate a structured data unit with rhythm feature expression capabilities.

[0019] As a preferred embodiment of the intelligent factory production supervision and management method described in the present invention, the standard semantic process flow diagram is a directed graph structure containing multiple task nodes and connection relationships. This is obtained by clustering the spectral vectors in the historical rhythm fluctuation spectrum and selecting the central spectrum as a representative.

[0020] As a preferred solution of the factory production intelligent supervision management method described in the present invention, the multi-dimensional semantic alignment detection compares the indicator semantic vector extracted from each node in the rhythm process nested graph with the standard semantic label corresponding to the node, determines the differences between the attributes, and outputs the residual measurement value of each node.

[0021] As a preferred solution of the factory production intelligent supervision and management method described in the present invention, the structured misalignment score list is a distributed structure constructed based on the alignment residual measurement results of each node, after aggregation and score sorting; the score threshold is set according to the historical process execution deviation range. When the misalignment score exceeds the score threshold, the node is identified as an abnormal node and the associated information is used to construct an abnormal behavior scenario graph.

[0022] In a second aspect, the present invention provides a factory production intelligent supervision and management system, comprising a primitive extraction module, a frequency domain transformation module, a deviation description module, a rhythm embedding module, a semantic alignment module, and a scene construction module;

[0023] The primitive extraction module is used to construct a time difference chain based on the operation behavior, extract the statistical characteristics of the time interval and the behavior trend within the window through a sliding window, and generate a fluctuation feature primitive that represents the operation rhythm; the frequency domain transformation module is used to perform frequency domain transformation on the fluctuation feature primitive, extract the frequency domain energy distribution characteristics and generate a rhythm fluctuation spectrum;

[0024] The deviation description module is used to calculate the dynamic time warping distance between the rhythm fluctuation spectrum and the historical rhythm reference spectrum to obtain a spectrum deviation descriptor;

[0025] The rhythm embedding module is used to assign deviation weight values, and nest the spectrum deviation descriptors according to the deviation weight values ​​and map them into the standard semantic process flow diagram to form a rhythm process nested diagram;

[0026] The semantic alignment module is used to extract node-level indicator semantic vectors from the rhythm process nested graph based on a graph embedding algorithm, perform multi-dimensional semantic alignment detection, and output an alignment residual metric;

[0027] The scenario construction module is used to aggregate alignment residual metrics, generate a structured misalignment score list, and construct an abnormal behavior scenario graph based on a score threshold.

[0028] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the factory production intelligent supervision management method as described in the first aspect of the present invention is implemented.

[0029] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the factory production intelligent supervision management method as described in the first aspect of the present invention.

[0030] The beneficial effects of the present invention are as follows: by constructing a time difference chain and extracting fluctuation feature primitives, combining frequency domain transformation to generate a rhythm fluctuation spectrum, and calculating the dynamic time regularization distance between it and the historical rhythm reference spectrum, it is possible to achieve modeling and deviation quantification of the rhythm of operational behavior, identify rhythm mutations and frequency anomalies, and improve the sensitivity and recognition accuracy of potential operational anomalies. By nesting and mapping the spectrum deviation descriptor to the standard semantic process flow chart, and extracting the node-level indicator semantic vector based on the graph embedding algorithm, performing multi-dimensional semantic alignment detection, and achieving structured comparison and deviation positioning between operational behavior and process semantics, the problem of difficulty in semantic interpretation and node traceability of operational behavior anomalies in the process path is solved. The ability to identify anomalies and semantic dislocations in the production supervision process is enhanced, and the perception depth of dynamic behavior and the structured expression ability of anomaly diagnosis are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0032] Figure 1 Provide an overall flow chart of the intelligent supervision and management method for factory production.

[0033] Figure 2 Generate a schematic diagram for the rhythmic fluctuation spectrum.

[0034] Figure 3 Schematic diagram of the rhythm process nesting process.

[0035] Figure 4 Construct a flow chart for the abnormal behavior scenario graph. DETAILED DESCRIPTION

[0036] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0037] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0038] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0039] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a method for intelligent supervision and management of factory production, including the following steps:

[0040] S1: Build a time difference chain based on the operation behavior, extract the statistical characteristics of the time interval and behavior trend within the window through the sliding window, and generate the fluctuation feature primitives that represent the operation rhythm.

[0041] Specifically, the following steps are included:

[0042] S1.1: Record the operating behaviors in the production process and form an operating behavior sequence.

[0043] Specifically, an operation record includes the timestamp of the operation, the operation type identifier, and the workstation number. The operation type identifier is used to distinguish specific operations, such as tightening, handling, testing, and confirmation; the workstation number is used to identify the physical location where the operation occurred. All records are sorted in ascending timestamp order to form an operation sequence.

[0044] S1.2: Calculate the time difference between adjacent operation behaviors based on the operation behavior sequence, and form a time difference chain based on the operation behavior type.

[0045] Specifically, based on the timestamps of adjacent operations in the action sequence, the time difference between them is calculated in seconds, forming a raw time interval sequence. The time difference between each pair of operations is then combined with the action type identifier (e.g., handling → tightening) and the corresponding workstation number to form a time difference data unit. All time difference data units are arranged in order of occurrence to form a time difference chain.

[0046] S1.3: Perform sliding window extraction on the time difference chain to extract continuous segments in the time difference chain.

[0047] Specifically, a fixed time window length and sliding step size are set to segment the time difference chain. The sliding window slides across the time difference chain in chronological order, extracting a fixed-length sequence of continuous time difference values ​​each time as a window segment. For example, if the time window length is set to 5 and the sliding step size is set to 1, each window contains 5 consecutive time difference data units. Sliding windows can overlap, covering the entire time difference chain, generating a set of window segments.

[0048] S1.4: Perform statistical calculations on the continuous segments in the time difference chain to obtain statistical indicators for expressing the operation rhythm characteristics.

[0049] Specifically, statistical analysis is performed on the time difference data units in each sliding window to extract statistical indicators. The statistical indicators include the time interval mean, time interval variance, time interval range, the diversity of operation behavior types within the window, and the trend of the mean before and after the window.

[0050] It should be noted that the time interval range refers to the difference between the maximum and minimum time intervals. The diversity of operation behavior types within a window refers to the number of different operation behavior combinations. The mean trend before and after a window refers to the difference from the mean of the previous window.

[0051] Statistical indicators are used to reflect the smoothness, volatility and trend changes of the rhythm of operating behavior.

[0052] S1.5: Combine statistical indicators into structured data units and combine them with operational behavior types to form fluctuation feature primitives.

[0053] Specifically, the statistical indicators are sequentially combined into a structured feature vector. This structured feature vector is then combined with the most frequently occurring action type label within the current sliding window to generate a fluctuation feature primitive. Each fluctuation feature primitive corresponds to a sliding window and has a fixed-length structure, action type identifier, and rhythmic fluctuation characteristics. This primitive is then used to generate a rhythmic fluctuation spectrum in the subsequent frequency domain analysis step.

[0054] Preferably, the time rhythm relationship between continuous operations can be accurately captured by timestamping and labeling the operation behaviors in the production process, and constructing a time difference chain; the time difference sequence is extracted in segments through a sliding window mechanism, and statistical indicators such as the mean, variance, range, operation behavior diversity and mean change trend of the time interval are calculated, which can comprehensively characterize the timing characteristics and rhythm fluctuations of the operation behavior; the statistical indicators are encoded into structured feature vectors and combined with the main operation behavior types to generate fluctuation feature primitives, so that the rhythm characteristics of the operation behavior can be input into the subsequent frequency domain processing process in a unified form, thereby realizing the modeling and quantification of irregular operation rhythms, laying the foundation for identifying potential rhythm anomalies in the production process, and improving the sensitivity and expression ability in complex behavior scenarios of the method.

[0055] S2: Perform frequency domain transformation on the fluctuation feature primitives, extract the frequency domain energy distribution characteristics and generate the rhythm fluctuation spectrum.

[0056] Specifically, the following steps are included:

[0057] S2.1: Arrange the fluctuation feature primitives in chronological order to form a fluctuation feature primitive sequence indexed by equally spaced time.

[0058] Specifically, the fluctuation feature primitives are sorted in ascending timestamp order. Each fluctuation feature primitive contains a structured feature vector derived from rhythm statistics extracted using a sliding window. The time index is determined by the sliding window start time, ensuring that the fluctuation feature primitives are continuous and evenly spaced along the time axis.

[0059] S2.2: Normalize the statistical indicators in the fluctuation characteristic primitive sequence.

[0060] Specifically, for each dimensional statistical indicator, the maximum and minimum values ​​within the entire sequence are extracted, and the original indicator values ​​are linearly mapped to the interval [0, 1] using the maximum and minimum value normalization method. Normalization improves the scale consistency of statistical indicators in each dimension and avoids interference caused by dimensional differences in subsequent frequency domain transformations.

[0061] S2.3: Use fast Fourier transform to perform frequency domain transformation on the fluctuation characteristic primitive sequence to obtain a spectrum representation.

[0062] Specifically, the frequency domain transform uses the Fast Fourier Transform (FFT) method to map the statistical indicator sequence in the time domain into a complex spectrum in the frequency domain. The sampling frequency is set according to the time index of the fluctuation characteristic primitive sequence, and the sampling rate is calculated based on the sampling frequency. A Hamming window function is used to reduce spectral leakage.

[0063] The normalized fluctuation characteristic primitive sequence is multiplied by the Hamming window function to perform windowing, and zero padding and FFT calculation are performed to obtain the amplitude spectrum, power spectrum and frequency axis.

[0064] S2.4: Extract frequency domain features based on the spectrum representation to construct a rhythm fluctuation spectrum.

[0065] Specifically, the frequency domain features include main frequency, high frequency energy ratio, spectrum kurtosis, spectrum variance and frequency band energy distribution.

[0066] The dominant frequency is obtained by selecting the frequency corresponding to the maximum amplitude in the amplitude spectrum.

[0067] The high-frequency energy ratio is obtained by setting the high-frequency threshold to 0.3 times the sampling rate and calculating the ratio of the sum of the power spectra exceeding the high-frequency threshold to the total power. The high-frequency threshold is set empirically and adjusted experimentally.

[0068] The spectral kurtosis is used to measure the sharpness of the spectrum peak, and the spectral variance reflects the degree of discreteness of the energy distribution.

[0069] Band energy distribution involves dividing the entire spectrum into low, mid, and high frequency bands, and calculating the energy contribution within each band. This spectrum division is based on the physical periodicity of the operating behavior and the spectrum analysis requirements of signal processing. The low frequency band is 0-0.01 Hz, the mid frequency band is 0.01-0.1 Hz, and the high frequency band is 0.1-0.2 Hz. The energy contribution of each band is calculated by summing the power spectra of all frequency points within the band and dividing it by the total power of the entire spectrum.

[0070] All frequency domain features are structured into JSON format files and saved as rhythm fluctuation spectra.

[0071] Preferably, by constructing the fluctuation feature primitives as a fluctuation feature primitive sequence with equally spaced time indexes and normalizing the fluctuation feature primitive sequence, the consistency and effectiveness of the rhythm statistical indicators of each dimension in the frequency domain analysis are ensured; by using the fast Fourier transform combined with the Hamming window function and the zero-padding operation, the operation behavior rhythm signal can be accurately converted from the time domain to the frequency domain, effectively suppressing spectrum leakage and improving spectrum clarity; in the frequency domain, multi-dimensional features such as the main frequency, high-frequency energy proportion, spectrum kurtosis, spectrum variance and sub-band energy distribution are extracted to construct a rhythm fluctuation spectrum, which can not only reveal the structural characteristics of the operation behavior rhythm in a fine-grained manner, but also sensitively capture hidden abnormal fluctuation patterns, thereby providing a highly expressive input basis for subsequent spectrum deviation recognition and rhythm anomaly analysis.

[0072] S3: Calculate the dynamic time warping distance between the rhythm fluctuation spectrum and the historical rhythm reference spectrum to obtain the spectrum deviation descriptor.

[0073] Specifically, the following steps are included:

[0074] S3.1: Extract frequency component sequences from the rhythm fluctuation spectrum and the historical rhythm reference spectrum respectively.

[0075] It should be noted that the historical rhythm reference spectrum is obtained by calculation from the database based on historical operation behaviors.

[0076] Structured spectrum data is extracted from the rhythm fluctuation spectrum and historical rhythm reference spectra. Each structured spectrum data contains the power value, amplitude, and other frequency domain features at the frequency axis and corresponding frequency. The frequency components on the frequency axis are extracted as the main analysis line to form a frequency component sequence. The frequency component sequence includes information such as the main frequency, high-frequency energy distribution points, and characteristic frequency points, which are used for subsequent alignment processing.

[0077] S3.2: The frequency component sequence of the rhythm fluctuation spectrum and the frequency component sequence of the historical rhythm reference spectrum are used as input, and the dynamic time warping algorithm is used to perform segment-by-segment alignment.

[0078] Specifically, the Dynamic Time Warping (DTW) algorithm is selected as the core alignment method. The frequency component sequence of the rhythm fluctuation spectrum is input as a one-dimensional time series, and the frequency component sequence of the historical rhythm reference spectrum is input as another sequence. The Dynamic Time Warping distance matrix is ​​constructed. The DTW algorithm minimizes the overall matching distance by finding the shortest path between the two frequency component sequences. The position index and corresponding amplitude information of each matched frequency point pair in the original sequence are retained for subsequent error calculation.

[0079] S3.3: After the dynamic time warping algorithm alignment is completed, the amplitude difference between each corresponding frequency component pair on the alignment path is obtained and accumulated to form a total distance value representing the degree of overall spectrum difference.

[0080] Specifically, each matched pair of frequency points in the DTW alignment path is traversed, and the absolute difference in amplitude between the two frequency points is calculated. All these differences are accumulated and summed in the alignment order to obtain a scalar total distance value. For example, if the alignment path contains 80 frequency point pairs, and the amplitude difference between each pair is 0.01, 0.03, 0.04, etc., then these 80 values ​​are accumulated to obtain the overall frequency domain structure offset of the rhythm fluctuation spectrum relative to the historical rhythm reference spectrum as the total distance value.

[0081] S3.4: Based on the total distance value, a combined feature vector is constructed in combination with the dynamic time warping path, and the output is a spectral deviation descriptor.

[0082] Specifically, while obtaining the total distance value, information from the alignment path is extracted as supplementary features to enrich the spectral deviation descriptor structure. Supplementary features include, but are not limited to, the length of the alignment path (i.e., the number of matched frequency point pairs), the maximum offset value on the frequency axis (i.e., the maximum jump between frequency indices), the standard deviation of the amplitude difference (a measure of the dispersion of the difference distribution), and the number of high amplitude offset segments (e.g., the number of frequency difference segments that continuously exceed a certain threshold).

[0083] The total distance value and the supplementary features are concatenated into a one-dimensional structured vector to form a spectral deviation descriptor. This spectral deviation descriptor represents the difference between the rhythm fluctuation spectrum and the historical rhythm reference spectrum and is used in the subsequent nested mapping step to complete the alignment analysis of the semantic process flow.

[0084] Optimally, highly sensitive identification of rhythmic changes in production operation behaviors is achieved by finely aligning and quantitatively analyzing the frequency domain differences between the rhythm fluctuation spectrum and the historical rhythm reference spectrum. Processing the frequency component sequence using a dynamic time warping algorithm not only effectively eliminates the effects of nonlinear mismatches in frequency domain changes, but also preserves the position information and amplitude characteristics of key frequency points. Furthermore, through the accumulation of amplitude differences and multidimensional alignment path statistics, a spectral deviation descriptor is constructed. This not only provides a quantitative indicator of the overall rhythmic deviation degree, but also captures the local structural characteristics of rhythmic disturbances, helping to improve the accuracy and robustness of the method in abnormal rhythm detection, semantic behavior mapping, and process monitoring.

[0085] S4: Assign deviation weight values, and nest and map the spectral deviation descriptors into the standard semantic process flow diagram according to the deviation weight values ​​to form a rhythm process nested diagram.

[0086] Specifically, the following steps are included:

[0087] S4.1: Parse node labels in the standard semantic process flow diagram and extract process stage identifiers and node attributes.

[0088] It should be noted that the standard semantic process flow diagram uses nodes to represent process stages and directed edges to represent the dependency order between process flows.

[0089] All nodes in the standard semantic process flow diagram are parsed sequentially to extract the process stage identifier and basic node attribute information, including the node's start and end timestamps, workstation number, and behavior label. Node time information is used to determine the corresponding relationship between spectral deviation descriptors on the time axis, and node behavior labels are used to assist in matching operation types.

[0090] S4.2: Perform time-process mapping on the spectral deviation descriptor.

[0091] Specifically, the spectral deviation descriptor is derived from the alignment of the rhythm fluctuation spectrum with the historical rhythm reference spectrum, retaining the time index. The time index in the spectral deviation descriptor is compared with the node time range in the standard semantic process flow diagram to find the covered node interval.

[0092] The mapping method is as follows: if the time index of the spectral deviation descriptor falls within the start and end time range of a node, the spectral deviation descriptor is mapped to the process stage represented by the node. If it spans multiple nodes, the overlap ratio of the spectral deviation descriptors in the time ranges of different nodes is calculated to form the deviation weight value.

[0093] S4.3: Assign deviation weights based on process stage duration.

[0094] Specifically, if a spectral deviation descriptor spans multiple process stages, a linear distribution is performed based on the time percentage in each process stage. For example, if a spectral deviation descriptor has a total duration of 60 seconds and spans two process stages A (30 seconds) and B (30 seconds), a 50% deviation weight is assigned to both stages A and B. If a process stage only contains a portion of the spectral deviation descriptor, the percentage is calculated based on the actual coverage time and weighted accordingly. The distribution result is represented as a set of key-value pairs: <process stage identifier, deviation weight value>.

[0095] S4.4: Write the spectral deviation descriptor into the corresponding node according to the deviation weight value to obtain a node set.

[0096] Specifically, the total distance value and supplementary features of each spectral deviation descriptor are associated with the corresponding process stage identifier and scaled by the deviation weight. For example, if the total distance value of the spectral deviation descriptor is 0.48 and the deviation weight corresponding to a node is 0.3, the deviation value written to that node is 0.48 × 0.3 = 0.144. All spectral deviation descriptors are written one by one into the nodes in the flow chart in this manner. Ultimately, each node contains a subset of spectral deviation features, forming a node set.

[0097] S4.5: Calculate the weights of directed edges based on the process dependencies and obtain the edge set.

[0098] Specifically, directed edges in a standard semantic process flow diagram reflect the logical or physical order between operational process stages. For any pair of adjacent nodes with a dependency relationship, a subset of spectral deviation descriptors is extracted from the node set. The weight of the directed edge is calculated by calculating the absolute value of the difference between the deviation values ​​of the pair of nodes. For example, if the deviation value of the preceding node is 0.144 and the deviation value of the succeeding node is 0.217, the edge weight is |0.217 - 0.144| = 0.073. This edge weight represents the degree of change in the rhythm anomaly from the previous stage to the next stage and is interpretable.

[0099] S4.6: Construct a rhythm-process nested graph based on the node set and edge set.

[0100] Preferably, the set of nodes containing spectral deviation features and the set of edges calculated based on process dependencies are combined into a graph structure. Each node in the graph structure carries a spectral deviation feature vector as an attribute, and each edge carries a weight value. This graph structure is a rhythm-process nested graph, reflecting the distribution and propagation trends of anomalies at the rhythm fluctuation level across different process stages. This can be used for subsequent tasks such as anomaly identification, rhythm optimization, or process reconstruction.

[0101] Optimally, by nesting and mapping spectral deviation descriptors with standard semantic process flow diagrams, a rhythm-process nested graph is formed, achieving precise alignment of rhythm fluctuation information with the process flow, enabling fine-grained localization of rhythm anomalies at the process node level. The structured representation of node deviation values ​​and edge weights not only clarifies the spatial propagation path of anomalies within the process flow but also enhances the interpretability and traceability of the mechanisms underlying rhythm anomalies. This constructed rhythm-process nested graph provides the data foundation and structural support for subsequent anomaly identification, process optimization, and decision support, enhancing the intelligent and refined level of factory production management.

[0102] S5: Extract node-level indicator semantic vectors from the rhythm process nested graph based on the graph embedding algorithm, perform multi-dimensional semantic alignment detection, and output alignment residual metrics.

[0103] Specifically, the following steps are included:

[0104] S5.1: Extract node attribute data from the rhythm process nested graph and construct the node feature matrix.

[0105] Specifically, each node in the rhythmic process nested graph contains a subset of spectral deviation features, including multiple dimensions such as total distance value, alignment path length, maximum frequency axis offset value, amplitude difference standard deviation, and the number of high-amplitude offset segments. Each of these spectral deviation feature dimensions is extracted and arranged in a uniform order to form a node feature vector. A two-dimensional node feature matrix is ​​constructed, using all nodes in the graph as samples and each node feature vector as a row. For example, if the graph contains 30 nodes and each node has a 5-dimensional spectral deviation feature, the node feature matrix size is 30×5.

[0106] S5.2: Use a two-layer graph convolutional network for feature propagation to generate node-level indicator semantic vectors.

[0107] Specifically, based on the graph structure information of the rhythm process nested graph and combined with the node feature matrix, it is input into the graph convolutional network for feature propagation.

[0108] The graph convolutional network adopts a two-layer stacked structure: the first layer of graph convolution linearly transforms and weighted aggregates the initial node features and first-order neighbor features to generate an intermediate semantic representation;

[0109] The second layer of graph convolution continues to propagate to a wider neighborhood based on the intermediate semantic representation, and outputs the final node-level indicator semantic vector.

[0110] Each layer of graph convolution performs a ReLU activation operation and uses a normalized adjacency matrix for structural weighting. For example, if the input node feature dimension is 5, the first layer output dimension is 16, and the second layer output dimension is 8, then each node ultimately generates an 8-dimensional node-level indicator semantic vector, representing the node's embedded semantic expression under the rhythm fluctuation deviation structure.

[0111] S5.3: Obtain the standard semantic label vector for each node from the standard semantic process flow graph.

[0112] Specifically, each node in the standard semantic process flow diagram has a process stage identifier and node attribute labels, including information such as the action type label, workstation number, operation duration, and historical expected cadence interval. These label data for each node are vectorized and encoded to form a structured standard semantic label vector.

[0113] The encoding methods include: one-hot encoding for behavior type, ordinal encoding for workstation number, and normalized numerical representation for operation duration and rhythm interval. The encoding results of each node are concatenated in a fixed order to form a standard semantic label vector.

[0114] S5.4: Perform multi-dimensional semantic alignment detection on the node-level indicator semantic vector and the standard semantic label vector, perform cosine similarity comparison and Euclidean distance residual calculation, and generate an alignment residual measurement set.

[0115] Specifically, for each node in the rhythm process nested graph, the corresponding node-level indicator semantic vector is compared with the standard semantic label vector of the node with the same identifier in the standard semantic process flow graph. The comparison method includes two indicators: cosine similarity comparison and Euclidean distance residual calculation:

[0116] Cosine similarity comparison: Calculates the cosine of the angle between two vectors, with a value range of [-1, 1], which is used to measure semantic direction consistency.

[0117] Euclidean distance residual calculation: Calculate the straight-line distance between two vectors in Euclidean space, which represents the overall residual degree of semantic expression.

[0118] The cosine similarity comparison is used as the output result of the alignment metric, the Euclidean distance residual is calculated as the output result of the residual metric, and saved as a structured indicator record. The cosine similarity and Euclidean residual results of all nodes constitute the alignment residual metric set.

[0119] Optimally, by nesting and comparing spectral deviation features with standard semantic labels through graph structure, a graph convolutional network is used to achieve deep extraction and structure-aware modeling of node-level indicator semantics. This in turn establishes a multidimensional semantic alignment mechanism between the rhythm process nested graph and the standard semantic process flow chart. Through a dual comparison method using cosine similarity and Euclidean distance, this method quantifies the consistency of node semantics and accurately assesses their residual offset, effectively enabling semantic diagnosis and quantitative expression of production rhythm anomalies, enhancing the intelligence, refinement, and interpretability of production process supervision and management.

[0120] S6: Aggregate the alignment residual metrics to generate a structured misalignment score list and construct an abnormal behavior scenario graph based on the score threshold.

[0121] Specifically, the following steps are included:

[0122] S6.1: Compute a misalignment score for each node based on a set of alignment residual metrics.

[0123] Specifically, the cosine similarity in the alignment residual metric set is transformed into cosine difference, and the Euclidean distance residual is normalized.

[0124] Taking the Euclidean distance residual as the basic linear term and adding cosine similarity as the weighting factor, the misalignment score calculation expression is:

[0125] S=α×D cos +β×R norm ;

[0126] Among them, S is the dislocation score, D cos is the cosine similarity transformed by cosine difference, R norm is the normalized Euclidean distance residual, α is the cosine similarity ratio, and β is the Euclidean distance residual ratio. α and β are initially set to 0.5 and adjusted according to the direction deviation and magnitude offset ratio to satisfy α + β = 1.

[0127] S6.2: Generate a structured misalignment score list based on the misalignment score.

[0128] It should be noted that the structured misalignment score list is sorted in descending order of misalignment score, and nodes with the same score are sorted in ascending order of timestamp.

[0129] S6.3: Set the scoring threshold based on the distribution range of node misalignment scores in the historical process flow.

[0130] Specifically, multiple groups of dislocation score lists are extracted from the historical semantic process, the historical score mean and historical score standard deviation are calculated, and the Z-Score method is used to set the score threshold as the historical score mean + 1.5×historical score standard deviation.

[0131] S6.4: Construct abnormal behavior scenario graph based on scoring threshold.

[0132] Specifically, the misalignment score list is traversed, and nodes whose misalignment scores exceed the score threshold are set as abnormal nodes. All abnormal nodes are extracted and integrated into an abnormal node set.

[0133] Based on the dependency edges in the standard semantic flow graph, only the edges connecting two abnormal nodes are retained to obtain the abnormal edge set.

[0134] An abnormal behavior scenario graph is constructed based on the abnormal node set and abnormal edge set, and the graph is drawn through the NetworkX library in Python to achieve visualization support.

[0135] Optimally, by constructing a structured misalignment score list and forming an abnormal behavior scenario graph, we achieve quantitative identification and centralized presentation of operational rhythm deviations and semantic misalignments. A comprehensive scoring of node alignment errors using cosine similarity and Euclidean distance residuals enhances the robustness and multi-dimensional expressiveness of misalignment assessment. By setting statistical scoring thresholds, we dynamically adapt to the needs of abnormality identification under different process contexts. The abnormal behavior scenario graph constructed based on highly misaligned nodes not only clarifies the structural distribution of abnormal behaviors but also supports intuitive location and correlation analysis of out-of-control links, thereby improving the accuracy and interpretability of anomaly identification in complex process scenarios.

[0136] This embodiment also provides a factory production intelligent supervision and management system, including: a primitive extraction module, a frequency domain transformation module, a deviation description module, a rhythm embedding module, a semantic alignment module and a scenario construction module;

[0137] The primitive extraction module is used to build a time difference chain based on the operation behavior, extract the statistical characteristics of the time interval and the behavior trend within the window through the sliding window, and generate the fluctuation feature primitives that represent the operation rhythm;

[0138] The frequency domain transformation module is used to perform frequency domain transformation on the fluctuation characteristic primitives, extract the frequency domain energy distribution characteristics and generate the rhythm fluctuation spectrum;

[0139] Deviation description module, used to calculate the dynamic time warping distance between the rhythm fluctuation spectrum and the historical rhythm reference spectrum to obtain the spectrum deviation descriptor;

[0140] The rhythm embedding module is used to assign deviation weight values ​​and embed the spectrum deviation descriptors into the standard semantic process flow diagram according to the deviation weight values ​​to form a rhythm process nested diagram;

[0141] The semantic alignment module is used to extract node-level indicator semantic vectors from the rhythm process nested graph based on the graph embedding algorithm, perform multi-dimensional semantic alignment detection, and output the alignment residual metric;

[0142] A scenario building module is used to aggregate alignment residual metrics, generate a structured misalignment score list, and construct an abnormal behavior scenario graph based on the score threshold.

[0143] This embodiment also provides a computer device suitable for the case of an intelligent supervision management method for factory production, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the intelligent supervision management method for factory production proposed in the above embodiment.

[0144] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0145] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for realizing intelligent supervision and management of factory production as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0146] In summary, the present invention can achieve modeling and deviation quantification of the rhythm of operational behavior, identify rhythm mutations and frequency anomalies, and improve the sensitivity and recognition accuracy of potential operational anomalies by: constructing a time difference chain and extracting fluctuation feature primitives, combining frequency domain transformation to generate a rhythm fluctuation spectrum, and calculating the dynamic time regularization distance between the rhythm fluctuation spectrum and the historical rhythm reference spectrum. By nesting and mapping the spectrum deviation descriptor to the standard semantic process flow chart, and extracting the node-level indicator semantic vector based on the graph embedding algorithm, performing multi-dimensional semantic alignment detection, and achieving structured comparison and deviation positioning between operational behavior and process semantics, the problem of difficult semantic interpretation and node traceability of operational behavior anomalies in the process path is solved. The ability to identify anomalies and semantic dislocations in the production supervision process is enhanced, and the perception depth of dynamic behavior and the structured expression ability of anomaly diagnosis are improved.

[0147] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for intelligent supervision and management of factory production, characterized by: include, Based on the operation behavior, a time difference chain is constructed, and the statistical characteristics and behavior trends of the time interval within the window are extracted through a sliding window to generate the fluctuation feature primitives that represent the operation rhythm; Perform frequency domain transformation on the fluctuation characteristic primitives, extract the frequency domain energy distribution characteristics and generate the rhythm fluctuation spectrum; Calculate the dynamic time warping distance between the rhythm fluctuation spectrum and the historical rhythm reference spectrum to obtain the spectrum deviation descriptor; Assign deviation weight values, and nest and map the spectrum deviation descriptors into the standard semantic process flow diagram according to the deviation weight values ​​to form a rhythm process nested diagram; Extract node-level indicator semantic vectors from the rhythm process nested graph based on graph embedding algorithm, perform multi-dimensional semantic alignment detection, and output alignment residual metrics; Aggregate the alignment residual metrics to generate a structured misalignment score list, and build an abnormal behavior scenario graph based on the score threshold.

2. The factory production intelligent supervision and management method according to claim 1, characterized in that: The spectrum deviation descriptor is used to characterize the overall difference between the rhythm fluctuation spectrum in the current window and the historical rhythm reference spectrum in terms of the main frequency position, high frequency energy distribution and spectrum morphology characteristics; Through the dynamic time warping algorithm, the frequency components of the rhythm fluctuation spectrum and the historical rhythm reference spectrum are aligned segment by segment, the cumulative difference values ​​of the corresponding frequency amplitudes on the alignment path are calculated, and the deviation feature vector containing time offset compensation is generated as the spectrum deviation descriptor based on the difference distribution.

3. The factory production intelligent supervision and management method according to claim 1, characterized in that: The specific steps of assigning deviation weight values ​​and mapping the spectrum deviation descriptors into the standard semantic process flow diagram according to the deviation weight values ​​to form a rhythm process nested diagram are as follows: Parse the node labels in the standard semantic process flow diagram and match the process stage identifiers associated with the spectral deviation descriptors; The deviation feature vector is assigned to the corresponding node according to the process stage weight, and the deviation value and timestamp association relationship are recorded in the node attributes to form a rhythm process nested graph.

4. The factory production intelligent supervision and management method according to claim 1, characterized in that: The time difference chain is composed of the time intervals between consecutive operation behaviors and is marked in combination with the operation behavior category and the workstation identification; The fluctuation feature primitive performs sliding extraction on the time difference chain by setting a fixed time window, and generates a structured data unit by combining the time interval mean, time interval variance, time interval range and the differential trend between adjacent windows within the window.

5. The factory production intelligent supervision and management method according to claim 1, characterized in that: The standard semantic process flow chart is a directed graph structure containing multiple task nodes and connection relationships. It is obtained by aggregating the spectrum vectors in the historical rhythm fluctuation spectrum using a clustering method and then selecting the central spectrum as a representative.

6. The factory production intelligent supervision and management method according to claim 1, characterized in that: The multi-dimensional semantic alignment detection compares the index semantic vector extracted from each node in the rhythm process nested graph with the standard semantic label corresponding to the node, determines the difference between the attributes, and outputs the residual measurement value of each node.

7. The factory production intelligent supervision and management method according to claim 1, characterized in that: The structured misalignment score list is a distributed structure constructed based on the alignment residual measurement results of each node, after aggregation and score sorting; the score threshold is set according to the historical process execution deviation range. When the misalignment score exceeds the score threshold, the node is identified as an abnormal node and the associated information is used to construct an abnormal behavior scenario graph.

8. A factory production intelligent supervision and management system, based on the factory production intelligent supervision and management method according to any one of claims 1 to 7, characterized in that: Including primitive extraction module, frequency domain transformation module, deviation description module, rhythm embedding module, semantic alignment module and scene construction module; The primitive extraction module is used to construct a time difference chain based on the operation behavior, extract the statistical characteristics of the time interval and the behavior trend within the window through the sliding window, and generate the fluctuation characteristic primitive that represents the operation rhythm; The frequency domain transformation module is used to perform frequency domain transformation on the fluctuation characteristic primitives, extract frequency domain energy distribution characteristics and generate a rhythm fluctuation spectrum; The deviation description module is used to calculate the dynamic time warping distance between the rhythm fluctuation spectrum and the historical rhythm reference spectrum to obtain a spectrum deviation descriptor; The rhythm embedding module is used to assign deviation weight values, and nest the spectrum deviation descriptors according to the deviation weight values ​​and map them into the standard semantic process flow diagram to form a rhythm process nested diagram; The semantic alignment module is used to extract node-level indicator semantic vectors from the rhythm process nested graph based on a graph embedding algorithm, perform multi-dimensional semantic alignment detection, and output an alignment residual metric; The scenario construction module is used to aggregate alignment residual metrics, generate a structured misalignment score list, and construct an abnormal behavior scenario graph based on a score threshold.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the factory production intelligent supervision and management method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the factory production intelligent supervision and management method according to any one of claims 1 to 7 are implemented.

Images