Manufacturing equipment intelligent operation monitoring method based on industrial vision

Abstract

The invention discloses a manufacturing equipment intelligent operation monitoring method based on industrial vision, and the method comprises the following steps: S1, collecting a continuous image sequence, and forming a visual semantic sequence; s2, inputting into an improved VMama model, introducing a running rhythm phase vector, executing phase conditional processing, and generating a running structure description; s3, constructing an operation semantic mode space; s4, generating an online operation structure description and an operation semantic evolution trajectory; s5, calculating a structure offset degree, executing accumulation judgment, and generating an operation abnormity identifier; s6, reversibility judgment is executed, and an operation risk level is generated; and S7, the operation semantic mode space is updated, and self-adaptive updating of the intelligent operation monitoring capability of the manufacturing equipment is completed. According to the invention, the continuous sensing of the operation state of the manufacturing equipment, the advanced recognition of the abnormal trend and the dynamic evaluation of the risk level are realized, and the technical effects of high monitoring precision, strong adaptability and good operation stability are achieved.

Description

Technical Field

[0001] This invention relates to the field of industrial automation and intelligent manufacturing technology, and in particular to an intelligent operation monitoring method for manufacturing equipment based on industrial vision. Background Technology

[0002] With the continuous improvement of intelligent manufacturing and industrial automation, the operation of manufacturing equipment under high-speed, continuous, and complex conditions has become the norm. How to continuously and accurately monitor the operating status of equipment has become a crucial technical issue to ensure production quality and operational safety. Industrial vision technology, due to its advantages such as non-contact operation, high information density, and adaptability to complex environments, is gradually being applied to the field of manufacturing equipment operation monitoring. Existing technologies typically acquire image sequences of the equipment's execution parts or processing areas and combine them with rule matching, statistical feature analysis, or traditional time series models to determine the equipment's operating status or detect anomalies.

[0003] However, in practical applications, existing methods for monitoring the operation of manufacturing equipment based on industrial vision still have significant shortcomings. On the one hand, existing technologies mostly analyze single-frame features or short-term features, lacking the ability to systematically model the rhythm, cyclic structure, and long-term temporal correlations of equipment operation. This results in unstable descriptions of the operating status under scenarios of changes in operating speed, fluctuations in cycle time, or switching of operating conditions, and the anomaly detection results are prone to misjudgment or omission. On the other hand, although some methods introduce temporal modeling structures, they usually do not distinguish and process the phase differences at different stages within the operating cycle, ignoring the structural differences of operating actions at different rhythmic positions, making it difficult for the generated operating features to accurately reflect the true structural changes of the equipment.

[0004] Furthermore, existing operational status assessments largely rely on fixed thresholds or static rules to identify anomalies, lacking an operational semantic pattern space based on historical stable operational states. This makes it impossible to continuously track and constrain the evolution trend of operational states, resulting in insufficient ability to identify slowly accumulating anomalies and reversible risks. When equipment operational states evolve, existing methods struggle to update operational status models in a timely manner, causing monitoring results to gradually deviate from the actual operational state, affecting the reliability and adaptability of long-term stable operation monitoring of manufacturing equipment.

[0005] Therefore, how to provide a method for intelligent operation monitoring of manufacturing equipment based on industrial vision is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an intelligent operation monitoring method for manufacturing equipment based on industrial vision. This invention collects continuous image sequences of the execution parts and processing areas of the manufacturing equipment, constructs a visual semantic sequence, and introduces the phase vector of the operation rhythm to perform phase-conditional temporal modeling of the visual semantic sequence to form an operation structure description. On this basis, an operation semantic pattern space is constructed, and online mapping, offset determination, reversibility discrimination, and risk level generation of the operation structure description are realized. Furthermore, the operation risk level drives the dynamic update of the operation semantic pattern space, realizing continuous learning and adaptive evolution of the operation status of the manufacturing equipment. This invention has the advantages of strong perception of changes in operation rhythm, high stability of anomaly identification, accurate operation risk classification, and outstanding adaptive capability for long-term operation monitoring.

[0007] An intelligent operation monitoring method for manufacturing equipment based on industrial vision according to an embodiment of the present invention includes the following steps: S1. Acquire a continuous sequence of images of the manufacturing equipment's execution parts and processing areas to form a visual semantic sequence; S2. Input the visual semantic sequence into the improved VMamba model, which introduces the running rhythm phase vector, performs phase conditionalization processing, and generates a running structure description. S3. Generate a historical operational structure description based on historical stable operational data, perform clustering processing to generate a set of operational semantic patterns, generate structural constraint intervals for the set of operational semantic patterns, and construct an operational semantic pattern space. S4. Generate an online operation structure description during the operation of the manufacturing equipment, map the online operation structure description to the operation semantic pattern space, and generate an operation semantic evolution trajectory. S5. Calculate the degree of structural offset between the online running structure description and the corresponding structural constraint interval in the running semantic pattern space based on the running semantic evolution trajectory, perform cumulative judgment, and generate running anomaly identifier; S6. Based on the operational anomaly identifier and the operational semantic evolution trajectory, perform reversibility judgment and generate operational risk level; S7. Update the operational semantic pattern space based on the operational structure description corresponding to the operational risk level to complete the adaptive update of the intelligent operation monitoring capability of manufacturing equipment.

[0008] Optionally, S1 specifically includes: By setting the sampling frame rate and acquisition time period of the industrial camera, continuous image acquisition is performed on the execution part and processing area of ​​the manufacturing equipment to obtain the original image sequence; The original image sequence is subjected to timestamp annotation and time alignment processing to obtain an aligned image sequence; Perform a periodic segmentation process on the aligned image sequence to obtain a periodic image sequence group; Perform running action segmentation processing on the periodic image sequence group to obtain the running action unit image sequence; Spatial location features, motion direction features, and rhythmic change features are extracted from the image sequence of the running action unit, and the spatial location features, motion direction features, and rhythmic change features are fused to generate the semantic vector of the running action unit; The semantic vectors of the action units are arranged according to the time sequence of the action units to form a visual semantic sequence.

[0009] Optionally, S2 specifically includes: Obtain the time index sequence corresponding to the semantic vector of the running action unit in the visual semantic sequence; Perform period normalization on the time index sequence to obtain the phase index sequence; The phase index sequence is subjected to embedding encoding to generate the running rhythm phase vector; The visual semantic sequence and the phase vector of the running rhythm are fed as joint inputs into the improved VMamba model; In the improved VMamba model, the running rhythm phase vector is introduced into the process of selecting scan parameters to perform phase conditional processing on the scan parameters corresponding to the visual semantic sequence; Temporal modeling is performed on the visual semantic sequence based on the scan parameters after phase conditionalization to generate a visual state sequence; Based on the visual state sequence, the temporal relationship between the running actions is constructed to generate a running structure description.

[0010] Optionally, the improved VMamba model specifically includes: The improved VMamba model includes input embedding processing, selective scan parameter generation processing, phase conditionalization processing, temporal modeling processing, and structural description generation processing. The input embedding process receives a visual semantic sequence and a running rhythm phase vector to generate a joint input representation. The running rhythm phase vector consists of a phase position component and a phase continuity component. The phase position component is obtained by periodic normalization of the time index sequence corresponding to the running action unit. The phase continuity component is obtained by mapping the phase position component through continuous values. The running rhythm phase vector is represented in vector form as a sequence including multiple continuous values. The selected scan parameter generation process generates scan parameters based on the joint input representation; The phase conditionalization process performs fusion constraints on the scan parameters and the phase vector of the running rhythm to generate phase conditional scan parameters; The temporal modeling process performs state sequence updates on the joint input representation based on phase-conditional scanning parameters to generate a visual state sequence. The structure description generation process generates a running structure description based on the visual state sequence.

[0011] Optionally, S3 specifically includes: Collect historical stable operation data and generate a set of historical operation structure descriptions based on the historical stable operation data; Clustering is performed on the historical operational structure description set to obtain the operational semantic pattern set, which includes pattern identifiers and corresponding pattern structure descriptions. The central structural description of the pattern is calculated based on the pattern structure description in the set of runtime semantic patterns; Calculate the structural distance between the historical operational structure description and the pattern center structure description in the historical operational structure description set, and form a structural distance set; The structural constraint interval is generated based on the set of structural distances. The structural constraint interval includes an upper bound and a lower bound of the structural distance. The runtime semantic pattern space is constructed based on the set of runtime semantic patterns and the structural constraint interval.

[0012] Optionally, S4 specifically includes: Collect continuous image sequences during the operation of manufacturing equipment and form online visual semantic sequences; Online visual semantic sequences are input into an improved VMamba model to generate an online running structure description; Calculate the structural distance between the online running structural description and the set of running semantic patterns in the running semantic pattern space, and generate a pattern distance sequence; Based on the pattern distance sequence, the operational semantic pattern identifier corresponding to the online operational structure description is determined, and an operational semantic mapping result is formed; The semantic mapping results of consecutive running cycles are arranged in the order of the running cycles to generate the semantic evolution trajectory of the running cycles.

[0013] Optionally, S5 specifically includes: Obtain the sequence of runtime semantic pattern identifiers from the runtime semantic evolution trajectory; Based on the sequence of runtime semantic pattern identifiers, obtain the structural constraint intervals in the runtime semantic pattern space corresponding to the runtime semantic pattern identifiers; Calculate the structural distance between the online running structural description and the pattern center structural description corresponding to the structural constraint interval to obtain the degree of structural offset; The structural offset degrees of consecutive operating cycles are arranged in the order of the operating cycles to generate an offset sequence; Perform a cumulative determination on the offset sequence. If the offset sequence meets the cumulative growth condition, generate a running error flag.

[0014] Optionally, if the offset sequence satisfies the cumulative growth condition, a running anomaly identifier is generated, specifically as follows: Perform differential calculation on the structural offset degree corresponding to adjacent running cycles in the offset sequence to obtain the offset change sequence; Perform a same-direction determination on the offset change sequence to obtain a continuously growing identifier sequence; The number of consecutive growth cycles is counted based on the continuous growth identifier sequence to form a growth duration parameter; The growth duration parameter is compared with the preset growth cycle threshold; if the growth duration parameter is greater than or equal to the growth cycle threshold, an operation anomaly flag is generated.

[0015] Optionally, S6 specifically includes: Obtain the runtime semantic evolution trajectory corresponding to the runtime anomaly identifier; extract the runtime semantic pattern identifier sequence before and after the anomaly occurs from the runtime semantic evolution trajectory; Calculate the sequence of structural distance changes between the runtime semantic pattern identifier and the runtime semantic pattern identifier before the anomaly occurred during consecutive runtime cycles. Regression judgment is performed on the structural distance change sequence to obtain the regression judgment result; reversibility judgment result is generated based on the regression judgment result; and operational risk level is generated based on the reversibility judgment result and the degree of structural offset.

[0016] Optionally, S7 specifically includes: Obtain the operational structure description corresponding to the operational risk level, and obtain the set of operational semantic patterns and the structural constraint interval in the operational semantic pattern space; Calculate the structural distance between the operational structure description and the pattern center structure description in the operational semantic pattern set, and generate a pattern distance sequence; Determine the runtime semantic pattern identifier corresponding to the runtime structure description based on the pattern distance sequence; If the pattern distance sequence is less than or equal to the upper bound of the structural distance of the structural constraint interval, the running structural description will be merged into the pattern structural description set corresponding to the running semantic pattern identifier, and the pattern center structural description will be updated. If the pattern distance sequence is greater than the upper bound of the structural distance of the structural constraint interval, the running structure description is added to the running semantic pattern set to generate a new running semantic pattern identifier, and a structural constraint interval is generated for the new running semantic pattern identifier. The runtime semantic pattern space is updated based on the updated set of runtime semantic patterns and the structural constraint interval.

[0017] The beneficial effects of this invention are: This invention addresses the problems of existing manufacturing equipment operation monitoring, which relies on single state thresholds, struggles to characterize operational rhythm changes, and lacks sufficient perception of gradual anomalies and structural evolution. It constructs a visual semantic sequence by collecting continuous image sequences of the execution parts and processing areas of the manufacturing equipment. An operational rhythm phase vector is introduced, and phase-conditional temporal modeling is performed on the visual semantic sequence within an improved VMamba model to generate an operational structure description with rhythm perception capabilities. Based on this, a set of operational semantic patterns and structural constraint intervals are constructed using historical stable operational data, forming an operational semantic pattern space. The online operational structure description is then mapped to this operational semantic pattern space to generate an operational semantic evolution trajectory. The structural offset within the operational semantic evolution trajectory is then analyzed. The invention achieves stable identification of operational anomalies through cumulative judgment of degree execution. Simultaneously, it performs reversibility discrimination by combining the structural distance changes of operational semantic pattern identifiers before and after the anomaly occurs, generating operational risk levels that distinguish between recoverable operational deviations and irreversible risk evolution. Finally, it dynamically updates the operational semantic pattern space based on the operational structure description corresponding to the operational risk level, enabling the operational semantic pattern space to continuously evolve with the long-term operation of the equipment. This invention realizes the structured expression of the operating status of manufacturing equipment, the identification of anomaly evolution trends, and the risk classification judgment without relying on fixed empirical thresholds. It significantly improves the stability of operational anomaly identification, the accuracy of operational risk assessment, and the adaptability and long-term effectiveness of intelligent operation monitoring capabilities of manufacturing equipment. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0019] Fig. 1 This is a flowchart of an intelligent operation monitoring method for manufacturing equipment based on industrial vision proposed in this invention; Fig. 2 This is a schematic diagram of the structure of the improved VMamba model proposed in this invention; Fig. 3 This is a data flow diagram of an intelligent operation monitoring method for manufacturing equipment based on industrial vision proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figs. 1-3 A method for intelligent operation monitoring of manufacturing equipment based on industrial vision includes the following steps: S1. Acquire a continuous sequence of images of the manufacturing equipment's execution parts and processing areas to form a visual semantic sequence; S2. Input the visual semantic sequence into the improved VMamba model, which introduces the running rhythm phase vector, performs phase conditionalization processing, and generates a running structure description. S3. Generate a historical operational structure description based on historical stable operational data, perform clustering processing to generate a set of operational semantic patterns, generate structural constraint intervals for the set of operational semantic patterns, and construct an operational semantic pattern space. S4. Generate an online operation structure description during the operation of the manufacturing equipment, map the online operation structure description to the operation semantic pattern space, and generate an operation semantic evolution trajectory. S5. Calculate the degree of structural offset between the online running structure description and the corresponding structural constraint interval in the running semantic pattern space based on the running semantic evolution trajectory, perform cumulative judgment, and generate running anomaly identifier; S6. Based on the operational anomaly identifier and the operational semantic evolution trajectory, perform reversibility judgment and generate operational risk level; S7. Update the operational semantic pattern space based on the operational structure description corresponding to the operational risk level to complete the adaptive update of the intelligent operation monitoring capability of manufacturing equipment.

[0022] This implementation method acquires continuous image sequences of the execution parts and processing areas of manufacturing equipment and constructs visual semantic sequences, transforming the operating status of the manufacturing equipment from discrete image information into a semantic expression with temporal consistency, thereby improving the completeness and stability of the operating status representation. Furthermore, the visual semantic sequence is input into an improved VMamba model of the operating rhythm phase vector, and a description of the operating structure is generated through phase conditional processing, enabling the temporal modeling process to have an operating cycle awareness capability, thus enhancing the accuracy of characterizing the periodic operating action structure of the manufacturing equipment. Simultaneously, an operating semantic pattern space is constructed based on historical stable operating data, allowing the normal operating structure of the manufacturing equipment to form quantifiable structural constraint boundaries, providing a unified reference for subsequent operating status determination. Furthermore, online operation... The structural description is mapped to the operational semantic pattern space to form an operational semantic evolution trajectory, enabling the operational state change process to have a continuous and traceable structural expression. Furthermore, by calculating the degree of structural offset between the operational structural description and the structural constraint interval and performing cumulative judgment, the gradual offset of the operational state is identified, thereby improving the detection capability of the hidden anomaly evolution process. At the same time, based on the operational anomaly identifier and the operational semantic evolution trajectory, reversibility discrimination is performed to establish a correspondence between the operational risk level and the structural offset evolution trend, thereby enhancing the accuracy of risk assessment. Furthermore, the operational semantic pattern space is updated based on the operational risk level, so that the operational semantic pattern space continuously evolves with the changes in the operational state of the manufacturing equipment, thereby achieving an adaptive improvement in the intelligent operation monitoring capability of the manufacturing equipment.

[0023] In this embodiment, S1 specifically refers to: During the operation of the manufacturing equipment, the sampling frame rate of the industrial camera is set to 20 to 60 frames per second, and the acquisition period covers the entire operating cycle of the manufacturing equipment. Continuous image acquisition is performed on the execution parts and processing areas of the manufacturing equipment to form a raw image sequence arranged in chronological order. Timestamp information is added to each frame of the original image sequence with millisecond precision. Time alignment processing is then performed on the original image sequence based on the timestamps to eliminate time deviations caused by multiple cameras or acquisition delays, thus forming a time-continuous and consistent aligned image sequence. Based on the periodic motion characteristics of the manufacturing equipment execution parts in the aligned image sequence, the aligned image sequence is subjected to operation cycle division processing. The operation cycle division is based on the completion of one complete repetitive operation of the manufacturing equipment execution parts as a cycle. Preferably, the length of the operation cycle is allowed to be adjusted within a 10% time fluctuation range, resulting in a group of periodic image sequences grouped by operation cycle. Within the periodic image sequence group, the running action segmentation process is performed on the periodic image sequence group according to the change of motion state of the execution part, dividing the continuous running cycle into multiple running action units. The running action unit corresponds to the stable motion stage or state switching stage of the manufacturing equipment within a running cycle, forming a running action unit image sequence. Spatial location features, motion direction features, and rhythmic variation features are extracted from each frame of the image sequence of the running action unit. Spatial location features are used to describe the position distribution of the execution part in the image coordinate system, motion direction features are used to describe the displacement direction of the execution part between consecutive frames, and rhythmic variation features are used to describe the changes in the motion velocity and acceleration of the execution part over time. The spatial location features, motion direction features, and rhythmic variation features are then fused. The fusion method is to stitch them together in chronological order and normalize them to generate a semantic vector of the running action unit. According to the chronological order of the operating action units within the operating cycle, the semantic vectors of the operating action units are arranged to form a visual semantic sequence with a clear temporal order, so that the visual semantic sequence can fully represent the continuous operating state changes of the manufacturing equipment within one or more operating cycles.

[0024] In this embodiment, S2 specifically refers to: Read the semantic vectors of running action units arranged in chronological order from the visual semantic sequence, and assign a corresponding time index to each semantic vector of running action units. The time index is represented in the form of consecutive integers, and the starting value corresponds to the start time of the running cycle, forming a time index sequence. Based on the results of the operation cycle division, the time index sequence is subjected to cycle normalization processing, which maps each time index to a relative position interval between zero and one. The cycle normalization processing is completed by dividing the time index by the total duration of the corresponding operation cycle, resulting in a phase index sequence that represents the relative positional relationship of the operation action within the operation cycle. The phase index sequence is subjected to embedding encoding processing to convert the phase index sequence into a multi-dimensional continuous numerical representation. Each phase index after embedding encoding consists of several continuous numerical values, which are used to simultaneously characterize the phase position and the phase change trend. Preferably, the dimension of the running rhythm phase vector is set to the range of eight to sixteen dimensions to generate the running rhythm phase vector. The visual semantic sequence and the phase vector of the running rhythm are aligned in the time dimension and concatenated according to the corresponding time index to form a joint input sequence. The joint input sequence is then input into the improved VMamba model. Inside the improved VMamba model, selection scan parameters are generated based on the joint input sequence. These selection scan parameters are used to control the state propagation range of the visual semantic sequence in the time dimension. During the generation process, a running rhythm phase vector is introduced, causing the selection scan parameters to change with the phase of the running cycle, thereby performing phase conditional processing on the scan parameters corresponding to the visual semantic sequence. Based on the scan parameters after phase conditional processing, temporal modeling is performed on the joint input sequence. During the temporal modeling process, the state update step size and update weight are controlled according to the scan parameters to generate a visual state sequence that reflects the evolution of visual semantics over time. Based on the state change relationship between adjacent time positions in the visual state sequence, the temporal correlation relationship between the operation actions is constructed. The temporal correlation relationship is expressed in a structured form to generate an operation structure description, so that the operation structure description fully reflects the action sequence, state continuity and rhythmic change characteristics of the manufacturing equipment in the operation cycle.

[0025] In this embodiment, the improved VMamba model specifically refers to: In the model input stage, the semantic vectors of the running action units arranged in chronological order in the visual semantic sequence are aligned with the phase vectors of the running rhythm in the time index dimension, and the semantic vectors and phase vectors at the corresponding positions are concatenated to form a joint input representation. The joint input representation is a multi-dimensional continuous numerical sequence containing visual semantic information and running rhythm information. The running rhythm phase vector is composed of a phase position component and a phase continuity component. The phase position component reflects the relative position of the running action unit in the running cycle, and the phase continuity component reflects the continuous trend of the phase position changing with time. Preferably, the sum of the dimensions of the visual semantic vector and the running rhythm phase vector in the joint input representation is controlled within the range of thirty-two to sixty-four dimensions. In the process of selecting scanning parameters and generating processing, scanning parameters are generated based on the numerical distribution of the joint input representation in the time dimension. The scanning parameters are used to describe the perception range of each time position to the preceding and following time positions during the state update process. The scanning parameters include scanning span parameters and scanning weight parameters. The scanning span parameters are used to limit the length of the reference time window during state update. Preferably, the time window length corresponding to the scanning span parameters is set to three to seven running action units. The scanning weight parameters are used to describe the relative contribution of different time positions in the state update. In the phase conditionalization stage, the running rhythm phase vector is introduced into the scanning parameter generation result, and the scanning span parameter and scanning weight parameter are subjected to fusion constraints, so that the scanning parameters change continuously with the phase of the running cycle. The fusion constraint method is to scale the scanning span parameter according to the phase position component and smoothly adjust the scanning weight parameter according to the phase continuous component, thereby forming phase conditional scanning parameters, so that the state update process has the running rhythm adaptive characteristics. In the temporal modeling stage, the joint input representation is updated with a state sequence based on the phase conditional scanning parameters. The state update process is performed by weighting and accumulating the joint input representation within the time window determined by the scanning span parameters, and assigning different update weights to the joint input representation at different time positions in combination with the scanning weight parameters, thereby generating a visual state sequence arranged in chronological order. Each state vector in the visual state sequence is a multi-dimensional continuous value used to describe the evolution of visual semantics within the running cycle. In the structural description generation and processing stage, based on the numerical change relationship between adjacent state vectors in the visual state sequence, the temporal correlation information between the running actions is extracted, and the temporal correlation information is expressed in a structured form, which includes the state sequence relationship and the state similarity relationship, thereby generating a running structural description that can fully reflect the action sequence, rhythm changes and state continuity characteristics of the manufacturing equipment during operation.

[0026] In this embodiment, the improved VMamba model is developed based on the VMamba model to address the issues of strong periodicity, strong rhythmicity, and gradual abnormal evolution in the operation of industrial vision scenarios for manufacturing equipment. Existing VMamba models focus on state-space modeling of visual sequences, with state updates primarily relying on the sequence content itself. They lack explicit constraints on the inherent periodic structure and rhythmic position information of the operation process. In scenarios of long-term continuous operation of manufacturing equipment, this can easily lead to insufficient differentiation of similar action states in different operating cycles, thus reducing the ability to recognize structural shifts. Therefore, this embodiment introduces a running rhythm phase vector into the scanning parameter generation path of the VMamba model. The relative temporal position of the running action within the operating cycle is used as conditional information in the state update control, enabling the model to maintain its original state-space modeling capabilities while possessing temporal modeling characteristics sensitive to the position of the operating cycle. By applying phase conditional constraints to the scanning parameters, the state propagation range and weights adaptively change with the running rhythm, allowing the same visual semantics to form differentiated structural expressions under different operating phases, thereby improving the ability of the operating structure description to represent periodic changes and gradual shifts. This implementation does not change the basic state space framework of the VMamba model. Instead, it introduces rhythmic condition information directly related to the industrial operation mechanism in the key parameter generation stage, realizing the deep integration of visual semantics, operation rhythm and state evolution. It has technical features that are different from existing time series modeling methods. It can enhance the perception of the evolution process of the manufacturing equipment operation structure without the need for additional supervision information, demonstrating clear novelty and creativity.

[0027] In this embodiment, S3 specifically refers to: During the long-term stable operation of the manufacturing equipment, the corresponding visual semantic sequences are collected, and the visual semantic sequences are input into the improved VMamba model to generate an operational structure description. The historical stable operation data is limited to a set of data with the lowest operational risk level and no less than one hundred consecutive operation cycles, thereby forming a set of historical operational structure descriptions. Based on the similarity relationship between the various operation structure descriptions in the historical operation structure description set, the historical operation structure description set is grouped so that operation structure descriptions with similar structural features are grouped into the same group. The grouping result forms an operation semantic pattern set. Each operation semantic pattern in the operation semantic pattern set corresponds to a pattern identifier and is associated with a group of pattern structure descriptions with similar structural features. Within the pattern structure description set corresponding to each running semantic pattern, the structural features of each dimension of the pattern structure description are numerically aggregated. By averaging the values ​​of the same structural feature in the pattern structure description set, a pattern center structure description that can represent the overall structural features of the running semantic pattern is generated. In the set of historical operation structure descriptions, the structural distance between each historical operation structure description and the pattern center structure description corresponding to its operation semantic mode is calculated. The structural distance is obtained by comprehensively considering the degree of difference between the operation structure description and the pattern center structure description in the corresponding structural feature dimension, forming a set of structural distances. Based on the structural distance set, statistical analysis is performed on the structural distance values. The minimum and maximum values ​​in the structural distance set are used as boundary references. On this basis, a safety margin is introduced to generate a structural constraint interval. The structural constraint interval includes a lower bound and an upper bound of the structural distance. Preferably, the lower bound of the structural distance is set as the minimum value of the structural distance set minus a 5% margin, and the upper bound of the structural distance is set as the maximum value of the structural distance set plus a 10% margin. Based on the set of operational semantic patterns and the corresponding structural constraint intervals, the pattern identifier, the pattern center structure description, and the structural constraint intervals are uniformly organized to construct an operational semantic pattern space, which can fully represent the typical operational structure distribution range of manufacturing equipment under stable operating conditions.

[0028] In this embodiment, S4 specifically refers to: During the actual operation of the manufacturing equipment, continuous image sequences of the execution part and the processing area are continuously collected, and an online visual semantic sequence is generated in accordance with the method described in the claims. The online visual semantic sequence is consistent with the visual semantic sequence formed in the historical stage in terms of semantic vector dimension and time indexing rules. The online visual semantic sequence is input into the improved VMamba model. By introducing the phase vector of the running rhythm and performing phase conditional processing, an online running structure description corresponding to the online running state is generated. The online running structure description expresses the structural relationship between running actions in the current running cycle in a multi-dimensional continuous numerical form. In the runtime semantic pattern space, the runtime semantic pattern set and the corresponding pattern center structure description are obtained, and the structural distance between the online runtime structure description and each pattern center structure description in the runtime semantic pattern set is calculated. The structural distance is obtained by combining the numerical differences between the online runtime structure description and the pattern center structure description in each structural feature dimension, forming a pattern distance sequence arranged according to the runtime semantic pattern. Based on the pattern distance sequence, the online running structure description is assigned to the running semantic pattern identifier with the smallest structure distance and within the corresponding structure constraint interval to form a running semantic mapping result. Preferably, when the minimum structure distance is less than or equal to the upper bound of the structure distance of the structure constraint interval, the corresponding running semantic pattern identifier is determined. The operational semantic mapping results obtained from consecutive operational cycles are arranged according to the time sequence of the operational cycles. The operational semantic pattern identifiers are then arranged into a sequence according to the time index to form an operational semantic evolution trajectory. The operational semantic evolution trajectory is used to characterize the structural change process of the manufacturing equipment's operating state within multiple operational cycles.

[0029] In this embodiment, S5 specifically refers to: Extract the sequence of operation semantic pattern identifiers arranged in the order of operation cycle from the operation semantic evolution trajectory. The sequence of operation semantic pattern identifiers reflects the changes in the operation semantic pattern of the manufacturing equipment in a continuous operation cycle. Based on the sequence of operation semantic pattern identifiers, the structural constraint intervals corresponding one-to-one with the operation semantic pattern identifiers are located in the operation semantic pattern space. The structural constraint intervals are jointly defined by the lower bound of the structural distance and the upper bound of the structural distance, and are used to describe the range of allowable fluctuations of the operation structure under the corresponding operation semantic pattern. Within each running cycle, the structural distance between the online running structural description and the pattern center structural description corresponding to the running semantic pattern identifier is calculated. The structural distance is obtained by comprehensively considering the numerical differences between the online running structural description and the pattern center structural description in each structural feature dimension. The structural distance is used as the degree of structural offset to quantify the degree of deviation of the current running structure from the stable pattern center. The structural offset degrees obtained from consecutive operating cycles are arranged according to the time sequence of the operating cycle to form an offset sequence that reflects the evolution of structural offset over time. The offset sequence represents the structural offset degree corresponding to each operating cycle in the form of continuous numerical values. The offset sequence is cumulatively determined based on the changing trend of the structural offset degree of adjacent operating cycles in the offset sequence. When the offset sequence shows a monotonically increasing characteristic in continuous operating cycles, it is determined to meet the cumulative growth condition. Preferably, the number of continuously increasing operating cycles is set to no less than three operating cycles. When the cumulative growth condition is met, an operating abnormality identifier is generated, so that the operating abnormality identifier corresponds to the state of continuous offset of the operating structure of the manufacturing equipment.

[0030] In this embodiment, generating an anomaly flag if the offset sequence satisfies the cumulative growth condition specifically means: Based on the degree of structural offset corresponding to adjacent operating cycles in the offset sequence, the difference between the degree of structural offset in the next operating cycle and the degree of structural offset in the previous operating cycle is calculated. The difference is obtained by subtracting the degree of structural offset from the degree of structural offset in the next operating cycle, forming an offset change sequence arranged in chronological order. The offset change sequence reflects the increase or decrease of structural offset in the form of continuous numerical values. Based on the sign of each difference in the offset change sequence, the same direction of the offset change sequence is determined. When the difference is positive, it is marked as a growth state, and when the difference is zero or negative, it is marked as a non-growth state. This forms a continuous growth identifier sequence composed of growth state and non-growth state. The continuous growth identifier sequence is used to describe the consistency of the structural offset change direction. In the continuous growth identifier sequence, the growth states that occur consecutively within adjacent operating cycles are statistically analyzed. The statistical results are represented by the number of consecutive growth cycles, which is used as a growth duration parameter to quantify the duration of continuous growth of structural offset. The growth duration parameter is compared with a preset growth cycle threshold, which is used to limit the minimum number of operating cycles required for the continuous growth of structural offset. Preferably, the growth cycle threshold is set to a range of three to five operating cycles. When the growth duration parameter is greater than or equal to the growth cycle threshold, an operation anomaly flag is generated, indicating that the corresponding manufacturing equipment operation structure has a continuous deviation and exceeds the stable fluctuation range.

[0031] In this embodiment, S6 specifically refers to: Based on the generated operation anomaly identifiers, locate the operation semantic evolution trajectory corresponding to the operation anomaly identifiers, and divide the operation stage before the anomaly occurs and the operation stage after the anomaly occurs in the operation semantic evolution trajectory according to the time index, and extract the operation semantic pattern identifier sequence before the anomaly occurs and the operation semantic pattern identifier sequence after the anomaly occurs. During the operation phase after the anomaly occurs, operation semantic pattern identifiers corresponding to multiple consecutive operation cycles are selected, and the structural distance change relationship between the operation semantic pattern identifiers and the operation semantic pattern identifiers corresponding to the last operation cycle before the anomaly occurs is calculated. The structural distance change relationship is composed of the difference between the structural distance of the subsequent operation cycle and the structural distance before the anomaly occurs, forming a structural distance change sequence arranged in the order of the operation cycles. Based on the structural distance change sequence, the trend of structural distance change is regressed and determined. The regression determination is completed by analyzing the directional characteristics of the changes in the values ​​of the structural distance change sequence with the running cycle. When the structural distance change sequence shows a continuous decreasing or stable downward trend, it is determined to be a regression state. When the structural distance change sequence shows a continuous increasing or no obvious downward trend, it is determined to be a non-regression state, thus forming the regression determination result. Based on the regression determination results, an invertibility discrimination result is generated, where the regression state corresponds to an invertible state and the non-regression state corresponds to an irreversible state. The operational risk is classified by combining the reversibility judgment result and the degree of structural deviation. When the reversibility judgment result is reversible and the degree of structural deviation is less than 1.20 times the upper limit of the structural constraint interval, the operational risk level is set to low risk. When the reversibility judgment result is reversible and the degree of structural deviation is greater than or equal to 1.20 times but less than 1.50 times the upper limit of the structural constraint interval, the operational risk level is set to medium risk. When the reversibility judgment result is irreversible or the degree of structural deviation is greater than or equal to 1.50 times the upper limit of the structural constraint interval, the operational risk level is set to high risk. This allows the operational risk level to reflect the severity and recovery possibility of the structural deviation of the manufacturing equipment.

[0032] In this embodiment, S7 specifically refers to: Obtain the operational structure description corresponding to the operational risk level, and simultaneously read the set of operational semantic patterns and the corresponding structural constraint intervals stored in the operational semantic pattern space. The operational structure description is represented in the form of a multi-dimensional continuous numerical vector. The vector dimension is consistent with the operational structure description generation stage. Each pattern structure description in the operational semantic pattern set is composed of a set of continuous numerical vectors of the same dimension. Based on the numerical difference between the operational structure description and the pattern center structure description in the operational semantic pattern set, the structural distance between the operational structure description and each pattern center structure description is calculated. The structural distance is obtained by squaring the difference between the operational structure description vector and the pattern center structure description vector in each dimension, summing them, and then taking the square root, forming a pattern distance sequence arranged in the order of the operational semantic pattern identifiers. Based on the runtime semantic pattern identifier corresponding to the minimum structure distance in the pattern distance sequence, candidate runtime semantic pattern identifiers for runtime structure description are determined, wherein the candidate runtime semantic pattern identifier is the pattern identifier with the smallest value in the pattern distance sequence; When the pattern distance corresponding to the candidate running semantic pattern identifier is less than or equal to the upper bound of the structure distance in the structure constraint interval, the running structure description is merged into the pattern structure description set corresponding to the candidate running semantic pattern identifier. The pattern center structure description is updated by taking the arithmetic mean of the pattern structure description set before merging and the newly added running structure description in each dimension. Preferably, the upper bound of the structure distance is the historical structure distance mean plus 1.30 times the standard deviation. When the pattern distance corresponding to the candidate running semantic pattern identifier is greater than the upper bound of the structural distance of the structural constraint interval, the running structure description is added to the running semantic pattern set as a new pattern structure description, and a new running semantic pattern identifier is generated. The new structural constraint interval is obtained by statistically analyzing the range of structural distance changes of the new pattern structure description within the initial running cycle. Preferably, the upper bound of the structural distance of the newly generated structural constraint interval is taken as 1.40 times the average of the initial structural distance, and the lower bound of the structural distance is taken as 0.80 times the average of the initial structural distance. Based on the updated set of operational semantic patterns and the corresponding structural constraint intervals, the operational semantic pattern space is replaced and updated as a whole, so that the operational semantic pattern space continues to evolve with the changes in the operational structure of the manufacturing equipment, thereby supporting the adaptive updating of the intelligent operation monitoring capabilities of the manufacturing equipment.

[0033] Example 1: To verify the feasibility of this invention in practice, it was applied to a high-speed CNC machining unit and automated loading / unloading station in an automotive parts factory. The manufacturing equipment included a vertical machining center, spindle fixtures, and a tool changer. The machining area exhibited typical on-site factors such as coolant splashing, metal chip reflection, fluctuations in lighting cycles, and cycle time switching. Such on-site conditions have long relied on manual inspections and single-threshold alarms. Common problems include frequent alarms when the cycle time slightly drifts, difficulty in early detection of progressive wear and minor fixture loosening, increased misjudgments during operating condition switching, and difficulty in distinguishing between recoverable deviations and irreversible risk evolution after anomalies occur, leading to unstable downtime for maintenance and quality fluctuations.

[0034] In this scenario, a continuous sequence of images of the manufacturing equipment's execution parts and processing area is simultaneously acquired by two industrial cameras. The sampling frame rate covers 30 to 50 frames per second, and the acquisition period covers the entire operating cycle, spanning tool changing and loading / unloading actions. After the image stream enters the process of this invention, it forms a visual semantic sequence according to the method described in the claims. The visual semantic sequence uses the semantic vector of the running action unit as the basic unit. The semantic vector is obtained by fusing spatial position features, motion direction features, and rhythmic change features, so that the states of the processing area such as "whether there is chip accumulation, whether the spindle feed is jittering, whether the fixture is in place, and whether there is lag in tool changing" are transformed from discrete images into continuous and calculable semantic expressions.

[0035] After the visual semantic sequence enters the improved VMamba model, a running rhythm phase vector is introduced and phase conditionalization processing is performed. The running rhythm phase vector is obtained from the time index sequence through periodic normalization and embedding encoding, which can explicitly represent the relative position of the running action in the running cycle and the continuous change trend of the phase. The improved VMamba model integrates the running rhythm phase vector in the selection of scan parameters, so that the scan span parameter and scan weight parameter are adaptively adjusted with the phase change, thereby avoiding confusion caused by the same visual semantic being "treated as the same state" under different running phases, and finally generating a running structure description sensitive to the differences in running stages. In this embodiment, this running structure description is used to express the strength of the temporal correlation and the changes in structural stability between "feed-cutting-retracting-tool changing-positioning-clamping-starting".

[0036] During the stable production phase, historical stable operation data from multiple consecutive days is collected to generate a historical operation structure description set. Clustering is then performed on this set to generate an operation semantic pattern set, and structural constraint intervals are generated for each set, constructing an operation semantic pattern space. In this embodiment, the operation semantic pattern space presents several typical patterns, such as normal cycle time, light load cycle time, and tool change intensive mode. Different patterns have their own pattern center structure description and upper and lower bounds for structural distance. During online operation of the manufacturing equipment, online operation structure descriptions are continuously generated and mapped to the operation semantic pattern space, forming an operation semantic evolution trajectory. When the degree of structural deviation meets the cumulative growth condition within a continuous operating cycle, an operation anomaly identifier is generated. The operation anomaly identifier triggers reversibility judgment. Regression judgment is performed based on the structural distance change sequence of the operation semantic pattern identifier sequence before and after the anomaly, and an operation risk level is generated accordingly, distinguishing between low-risk recoverable deviations, medium-risk requiring planned maintenance, and high-risk requiring immediate action. Finally, the operational structure description corresponding to the operational risk level drives the update of the operational semantic pattern space in reverse, so that the operational semantic pattern space continues to evolve with changes in tool batches, lubrication status, and shift operation differences, avoiding monitoring drift caused by the rigidity of pattern boundaries after long-term operation.

[0037] For ease of comparison, this embodiment sets up a control group using traditional industrial visual thresholding and short-window temporal statistical methods, mainly employing frame-level feature thresholding, fixed window mean and variance, and fixed alarm thresholds; the experimental group uses the method of this invention. Over a three-month period, production logs, equipment maintenance records, quality traceability, and manual verification results were used as the basis for labeling. Abnormal events included progressive or phased anomalies such as slight fixture loosening, tool change delays, slight spindle wobble, chip accumulation and obstruction, and coolant nozzle misalignment.

[0038] The table below shows a comparison of the performance of anomaly identification over three months. The detection rate refers to the proportion of abnormal events that were manually verified and confirmed by the system and were identified and generated as an anomaly. The false alarm rate refers to the proportion of anomaly indicators generated by the system but which were found to be normal fluctuations upon verification. The average warning lead time refers to the average number of minutes from the first issuance of an anomaly indicator to on-site confirmation and handling. The average judgment delay refers to the average calculation delay for a single operating cycle from visual semantic sequence to output of an anomaly indicator.

[0039] Table 1. Comparison of Anomaly Detection Results over Three Months

[0040] Table 1 shows that the stability of the present invention in identifying progressive anomalies under complex field conditions is significantly better than that of the control group. The detection rate of the control group showed a downward trend within three months, mainly because the fixed threshold and short window statistics are more sensitive to appearance disturbances caused by cycle time fluctuations, illumination cycle fluctuations, and coolant splashing. As tool wear and shift operation differences accumulate, the normal fluctuation distribution gradually drifts, and the fixed threshold cannot be adjusted synchronously, resulting in an increase in missed detections and false alarms. The present invention introduces the operating rhythm phase vector through an improved VMamba model and performs phase conditional processing, so that the same visual semantics form differentiated structural expressions in different operating phases, which can reduce the confusion caused by the difference between different stages within the cycle when switching operating conditions. At the same time, the operating semantic pattern space provides a structural constraint interval as a reference, and the degree of structural deviation adopts the cumulative judgment mechanism of continuous operating cycles, which is more in line with the abnormal evolution law of "progressive deviation". Therefore, the detection rate is consistently maintained above 90% and the false alarm rate is kept at a low level. In terms of early warning lead time, the present invention can give an anomaly mark when the structural deviation just appears and continues to increase, so that maintenance personnel can complete the handling before the fixture loosening and tool change delay trigger the shutdown, and the early warning lead time is significantly improved. The judgment delay increased slightly but remained within the range that could meet the requirements of online monitoring, and did not affect the cycle operation.

[0041] Regarding operational risk levels, this embodiment classifies operational risk into low, medium, and high risk, using manually verified "recoverable deviations" and "irreversible risk evolution" as the criteria for judgment. Recoverable deviations include recovery from short-term coolant blockage, recovery after cleaning temporary chip accumulation, and regression from minor cycle time disturbances; irreversible risk evolution includes the continued aggravation of fixture loosening, the increased tendency of spindle runout, and the continued expansion of hysteresis due to tool changer wear. The table below compares the accuracy of risk classification, the accuracy of reversibility judgment, and the drift suppression effect brought about by the update of the operational semantic pattern space over three months, and also shows the changes in the number of production-related downtime maintenance and the defect rate, to reflect the comprehensive benefits of monitoring results on production.

[0042] Table 2 Comparison of Risk Classification and Adaptive Update Effects over Three Months

[0043] Table 2 demonstrates that the key aspect of this invention, "how to handle anomalies," is closer to real-world production needs. The control group typically only provides binary alarms indicating whether an anomaly is present, lacking the ability to structurally discriminate anomaly regression trends. On-site, conservative shutdowns for investigation are often adopted, leading to a monthly increase in unplanned maintenance shutdowns. Simultaneously, it struggles to distinguish between recoverable deviations and irreversible risk evolution, resulting in low accuracy in risk classification and reversibility discrimination. This invention, after generating operational anomaly identifiers, further performs reversibility discrimination based on the operational semantic evolution trajectory. Regression is performed using the structural distance change sequence of the operational semantic pattern identifier sequence before and after the anomaly occurs. This allows low-risk recoverable deviations to be closed-loop through rapid on-site handling, while medium- and high-risk deviations trigger planned maintenance or immediate handling, thereby reducing unnecessary downtime. The adaptive update of the operational semantic pattern space brings a significant drift suppression effect. The decrease in false alarms continues to expand after the update, indicating that the pattern center structure description and structural constraint interval can evolve synchronously with long-term operational changes, preventing fixed thresholds from gradually becoming ineffective after tool batch changes, lubrication status changes, and accumulated operational differences. In terms of comprehensive indicators, the number of unplanned downtime maintenance decreased, and the quality defect rate decreased accordingly. This indicates that the operational risk level output by this invention can guide more reasonable maintenance timing and handling strategies, and realize the direct benefit of monitoring results to production stability and quality stability.

[0044] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent operation monitoring of manufacturing equipment based on industrial vision, characterized in that, Includes the following steps: S1. Acquire a continuous sequence of images of the manufacturing equipment's execution parts and processing areas to form a visual semantic sequence; S2. Input the visual semantic sequence into the improved VMamba model, which introduces the running rhythm phase vector, performs phase conditionalization processing, and generates a running structure description. S3. Generate a historical operational structure description based on historical stable operational data, perform clustering processing to generate a set of operational semantic patterns, generate structural constraint intervals for the set of operational semantic patterns, and construct an operational semantic pattern space. S4. Generate an online operation structure description during the operation of the manufacturing equipment, map the online operation structure description to the operation semantic pattern space, and generate an operation semantic evolution trajectory. S5. Calculate the degree of structural offset between the online running structure description and the corresponding structural constraint interval in the running semantic pattern space based on the running semantic evolution trajectory, perform cumulative judgment, and generate running anomaly identifier; S6. Based on the operational anomaly identifier and the operational semantic evolution trajectory, perform reversibility judgment and generate operational risk level; S7. Update the operational semantic pattern space based on the operational structure description corresponding to the operational risk level to complete the adaptive update of the intelligent operation monitoring capability of manufacturing equipment.

2. The intelligent operation monitoring method for manufacturing equipment based on industrial vision according to claim 1, characterized in that, Specifically, S1 is: By setting the sampling frame rate and acquisition time period of the industrial camera, continuous image acquisition is performed on the execution part and processing area of ​​the manufacturing equipment to obtain the original image sequence; The original image sequence is subjected to timestamp annotation and time alignment processing to obtain an aligned image sequence; Perform a periodic segmentation process on the aligned image sequence to obtain a periodic image sequence group; Perform running action segmentation processing on the periodic image sequence group to obtain the running action unit image sequence; Spatial location features, motion direction features, and rhythmic change features are extracted from the image sequence of the running action unit, and the spatial location features, motion direction features, and rhythmic change features are fused to generate the semantic vector of the running action unit; The semantic vectors of the action units are arranged according to the time sequence of the action units to form a visual semantic sequence.

3. The intelligent operation monitoring method for manufacturing equipment based on industrial vision according to claim 1, characterized in that, Specifically, S2 is: Obtain the time index sequence corresponding to the semantic vector of the running action unit in the visual semantic sequence; Perform period normalization on the time index sequence to obtain the phase index sequence; The phase index sequence is subjected to embedding encoding to generate the running rhythm phase vector; The visual semantic sequence and the phase vector of the running rhythm are fed as joint inputs into the improved VMamba model; In the improved VMamba model, the running rhythm phase vector is introduced into the process of selecting scan parameters to perform phase conditional processing on the scan parameters corresponding to the visual semantic sequence; Temporal modeling is performed on the visual semantic sequence based on the scan parameters after phase conditionalization to generate a visual state sequence; Based on the visual state sequence, the temporal relationship between the running actions is constructed to generate a running structure description.

4. The intelligent operation monitoring method for manufacturing equipment based on industrial vision according to claim 1, characterized in that, The improved VMamba model is specifically as follows: The improved VMamba model includes input embedding processing, selective scan parameter generation processing, phase conditionalization processing, temporal modeling processing, and structural description generation processing. The input embedding process receives a visual semantic sequence and a running rhythm phase vector to generate a joint input representation; The operating rhythm phase vector consists of a phase position component and a phase continuity component. The phase position component is obtained by periodic normalization of the time index sequence corresponding to the operating action unit. The phase continuity component is obtained by continuous numerical mapping of the phase position component. The operating rhythm phase vector is represented in vector form as a sequence including multiple continuous values. The selected scan parameter generation process generates scan parameters based on the joint input representation; The phase conditionalization process performs fusion constraints on the scan parameters and the phase vector of the running rhythm to generate phase conditional scan parameters; The temporal modeling process performs state sequence updates on the joint input representation based on phase-conditional scanning parameters to generate a visual state sequence. The structure description generation process generates a running structure description based on the visual state sequence.

5. The intelligent operation monitoring method for manufacturing equipment based on industrial vision according to claim 1, characterized in that, Specifically, S3 is: Collect historical stable operation data and generate a set of historical operation structure descriptions based on the historical stable operation data; Clustering is performed on the historical operational structure description set to obtain the operational semantic pattern set, which includes pattern identifiers and corresponding pattern structure descriptions. The central structural description of the pattern is calculated based on the pattern structure description in the set of runtime semantic patterns; Calculate the structural distance between the historical operational structure description and the pattern center structure description in the historical operational structure description set, and form a structural distance set; The structural constraint interval is generated based on the set of structural distances. The structural constraint interval includes an upper bound and a lower bound of the structural distance. The runtime semantic pattern space is constructed based on the set of runtime semantic patterns and the structural constraint interval.

6. The intelligent operation monitoring method for manufacturing equipment based on industrial vision according to claim 1, characterized in that, Specifically, S4 is: Collect continuous image sequences during the operation of manufacturing equipment and form online visual semantic sequences; Online visual semantic sequences are input into an improved VMamba model to generate an online running structure description; Calculate the structural distance between the online running structural description and the set of running semantic patterns in the running semantic pattern space, and generate a pattern distance sequence; Based on the pattern distance sequence, the operational semantic pattern identifier corresponding to the online operational structure description is determined, and an operational semantic mapping result is formed; The semantic mapping results of consecutive running cycles are arranged in the order of the running cycles to generate the semantic evolution trajectory of the running cycles.

7. The intelligent operation monitoring method for manufacturing equipment based on industrial vision according to claim 1, characterized in that, Specifically, S5 is: Obtain the sequence of runtime semantic pattern identifiers from the runtime semantic evolution trajectory; Based on the sequence of runtime semantic pattern identifiers, obtain the structural constraint intervals in the runtime semantic pattern space corresponding to the runtime semantic pattern identifiers; Calculate the structural distance between the online running structural description and the pattern center structural description corresponding to the structural constraint interval to obtain the degree of structural offset; The structural offset degrees of consecutive operating cycles are arranged in the order of the operating cycles to generate an offset sequence; Perform a cumulative determination on the offset sequence. If the offset sequence meets the cumulative growth condition, generate a running error flag.

8. The intelligent operation monitoring method for manufacturing equipment based on industrial vision according to claim 7, characterized in that, If the offset sequence meets the cumulative growth condition, an operation anomaly flag is generated, specifically as follows: Perform differential calculation on the structural offset degree corresponding to adjacent running cycles in the offset sequence to obtain the offset change sequence; Perform a same-direction determination on the offset change sequence to obtain a continuously growing identifier sequence; The number of consecutive growth cycles is counted based on the continuous growth identifier sequence to form a growth duration parameter; The growth duration parameter is compared with the preset growth cycle threshold; if the growth duration parameter is greater than or equal to the growth cycle threshold, an operation anomaly flag is generated.

9. The intelligent operation monitoring method for manufacturing equipment based on industrial vision according to claim 1, characterized in that, Specifically, S6 is: Obtain the runtime semantic evolution trajectory corresponding to the runtime anomaly identifier; extract the runtime semantic pattern identifier sequence before and after the anomaly occurs from the runtime semantic evolution trajectory; Calculate the sequence of structural distance changes between the runtime semantic pattern identifier and the runtime semantic pattern identifier before the anomaly occurred during consecutive runtime cycles. Regression analysis was performed on the structural distance change sequence to obtain the regression results; Reversibility determination results are generated based on the regression results; The operational risk level is generated based on the reversibility assessment results and the degree of structural offset.

10. The intelligent operation monitoring method for manufacturing equipment based on industrial vision according to claim 1, characterized in that, Specifically, S7 is: Obtain the operational structure description corresponding to the operational risk level, and obtain the set of operational semantic patterns and the structural constraint interval in the operational semantic pattern space; Calculate the structural distance between the operational structure description and the pattern center structure description in the operational semantic pattern set, and generate a pattern distance sequence; Determine the runtime semantic pattern identifier corresponding to the runtime structure description based on the pattern distance sequence; If the pattern distance sequence is less than or equal to the upper bound of the structural distance of the structural constraint interval, the running structural description will be merged into the pattern structural description set corresponding to the running semantic pattern identifier, and the pattern center structural description will be updated. If the pattern distance sequence is greater than the upper bound of the structural distance of the structural constraint interval, the running structure description is added to the running semantic pattern set to generate a new running semantic pattern identifier, and a structural constraint interval is generated for the new running semantic pattern identifier. The runtime semantic pattern space is updated based on the updated set of runtime semantic patterns and the structural constraint interval.

Images