Method for automatically detecting surface defects of printed matter by utilizing machine vision

Abstract

The invention discloses a method for automatically detecting surface flaws of printed matters by utilizing machine vision, which relates to the technical field of printing flaw detection and comprises the following steps of: synchronously acquiring data in a printing process to form a multi-modal time sequence data set; key dynamic features are extracted and quantified from the multi-modal time sequence data set, and a unified system state vector is generated; constructing a system state stability evolution model, inputting the system state vector into the stability evolution model, and calculating and outputting a real-time system stability index; in response to the fact that the real-time system stability index is lower than a preset early warning threshold value, causal decoupling analysis is started, and a current dominant failure path is judged; and according to the dominant failure path, generating a differential intervention instruction. According to the method, the transformation from passive fault response to active trend prediction, from fuzzy fault diagnosis to accurate failure attribution and from universal intervention to differential intelligent regulation and control is realized, the product quality is guaranteed on the premise of not interrupting production, and the production efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of printing defect detection, and in particular to a method for automatically detecting surface defects of printed products using machine vision. Background Art

[0002] High-speed web-to-web printing is key to large-scale, efficient production, and its product quality demands high system stability. During the printing process, the volatility of ink solvents can affect the drying and spreading of the product, and micro-vibrations generated by high-speed equipment can affect ink transfer and substrate transport.

[0003] Existing technologies usually use independent monitoring systems. Such monitoring systems not only have a delayed response and only issue alarms after defects occur on a large scale, which can easily lead to a large amount of waste, but also make it difficult to quickly and accurately determine the root cause of the problem. In addition, there are also problems such as a single intervention strategy, low efficiency, possible introduction of new disturbances, and inability to effectively identify and decouple critical system transitions triggered by the coupling of chemical and physical factors. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for automatically detecting surface defects of printed matter using machine vision, which solves the problems existing in the background technology.

[0005] To solve the above technical problems, the present invention provides a method for automatically detecting surface defects of printed matter using machine vision, comprising the following steps:

[0006] S1. Synchronously collect time-series defect image data, key component vibration data, and substrate surface temperature data during the printing process to form a multimodal time-series dataset;

[0007] S2. Extracting and quantifying key dynamic features from the multimodal time series dataset to generate a unified system state vector; the key dynamic features include visual features, vibration features, and thermodynamic features;

[0008] S3. Constructing a system state stability evolution model, inputting the system state vector into the stability evolution model, and calculating and outputting a real-time system stability index; the stability evolution model is constructed based on the deviation between the system state vector and a preset reference stable state vector;

[0009] S4. In response to the real-time system stability index being lower than a preset warning threshold, initiating causal decoupling analysis, calculating a chemical dominant contribution and a physical dominant contribution based on the system state vector, and determining a current dominant failure path by comparing the chemical dominant contribution and the physical dominant contribution;

[0010] S5. Generate differentiated intervention instructions based on the dominant failure path; if the dominant failure path is determined to be chemically dominant, generate adjustment instructions for the ink mixing system; if the dominant failure path is determined to be physically dominant, generate early warning instructions for the equipment maintenance system.

[0011] Preferably, the S2 specifically includes:

[0012] S21, based on the temporal defect image data, extracting and combining visual feature sub-vectors by calculating image blur, texture entropy, and color gradient norm;

[0013] S22. Based on the vibration data of the key components, extract the energy amplitude at the relevant characteristic frequency of the key components by performing short-time Fourier transform on the vibration data of the key components, and combine them into a vibration characteristic subvector;

[0014] S23, based on the substrate surface temperature data, by calculating the surface average temperature, temperature standard deviation and surface maximum temperature gradient, extracting and combining them into a thermodynamic characteristic subvector;

[0015] S24 , normalizing the visual feature sub-vector, the vibration feature sub-vector, and the thermodynamic feature sub-vector and fusing them into the unified system state vector.

[0016] Preferably, the S3 specifically includes:

[0017] S31. Calculate the difference between the system state vector and the preset reference stable state vector to obtain a system state deviation vector;

[0018] S32, performing a weighted quadratic calculation of a preset weight matrix on the system state deviation vector to generate a system potential energy value representing the degree to which the system deviates from a stable point in the potential energy field;

[0019] S33. Based on the time series of the system potential energy value, calculate the product of the time series autocorrelation and variance of the system potential energy value, and obtain the inverse of the product to obtain the real-time system stability index.

[0020] Preferably, the reference steady-state vector is obtained by continuously collecting and analyzing printing process data with a preset yield rate higher than a specific standard, and calculating the time average value of the calculated system state vector at each moment;

[0021] The preset weight matrix is optimized by analyzing historical failure data using Lasso regression to quantitatively determine the contribution of each key dynamic feature to the final product quality defects.

[0022] Preferably, the S4 specifically includes:

[0023] S41, decoupling the system state vector and the preset weight matrix into a chemical part and a physical part, respectively, wherein the chemical part corresponds to visual and thermodynamic characteristics related to ink, and the physical part corresponds to vibration characteristics related to equipment;

[0024] S42, calculating the chemical dominant contribution based on the system state vector, the reference stable state vector and the weight matrix of the chemical part;

[0025] S43, calculating the physical dominant contribution based on the system state vector, the reference stable state vector and the weight matrix of the physical part;

[0026] S44, calculating the ratio of the chemical dominant contribution to the physical dominant contribution to obtain a dominant pathway ratio;

[0027] S45. If the dominant pathway ratio is higher than the preset chemical-dominant decision threshold, the dominant failure pathway is determined to be chemically dominant; if the dominant pathway ratio is lower than the preset physical-dominant decision threshold, the dominant failure pathway is determined to be physically dominant.

[0028] Preferably, the chemical-dominant decision threshold and the physical-dominant decision threshold are calibrated through a controlled fault injection experiment, specifically including:

[0029] a) introducing a perturbation of excessively fast solvent evaporation into the ink system during the experiment, continuously recording and analyzing the statistical distribution of the dominant pathway ratio under this condition, so as to determine the chemical dominance decision threshold;

[0030] b) By applying known weak mechanical vibration excitation to key rotating components in the experiment, the statistical distribution of the dominant path ratio under this condition is continuously recorded and analyzed to determine the physical dominant decision threshold.

[0031] Preferably, the preset warning threshold is determined by statistically analyzing a large amount of system data transitioning from normal state to fault state, and based on receiver operating characteristic curve analysis, selecting a corresponding value that can maximize the warning recall rate while ensuring an acceptable false positive rate.

[0032] Preferably, the S5 specifically includes:

[0033] S51. If the dominant failure path is determined to be chemically dominant, a command is automatically generated and sent to the ink viscosity control system to adjust the solvent replenishment parameters, and a prompt message indicating an abnormal ink status is pushed to the human-computer interaction interface;

[0034] S52: If the dominant failure path is determined to be physically dominant, then based on the characteristic frequency information in the vibration characteristic subvector, the specific abnormal vibration component is identified, and a high-priority preventive maintenance work order for the component is automatically generated.

[0035] A system for automatically detecting surface defects of printed matter using machine vision is also provided, comprising:

[0036] Multimodal data synchronous acquisition module, used to synchronously collect time-series defect image data, key component vibration data and substrate surface temperature data during the printing process to form a multimodal time-series data set;

[0037] A dynamic feature vector construction module is used to extract and quantify key dynamic features reflecting the system state from the multimodal time series data set and fuse them into a unified system state vector;

[0038] A system state stability evolution modeling module is used to construct a model describing the dynamic evolution of the system state vector based on the system state vector, and calculate and output a real-time system stability index based on the model;

[0039] a dominant failure path causal decoupling module, configured to be activated in response to a condition where the real-time system stability index falls below a preset warning threshold to analyze the system state vector, thereby decoupling and determining the chemically dominant or physically dominant failure path that causes system instability;

[0040] The precise differentiated intervention decision module is used to generate and output differentiated control instructions for the chemical pathway or the physical pathway according to the dominant failure pathway determined by the causal decoupling module.

[0041] Beneficial effects

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] 1. By constructing an evolutionary model that can proactively assess system stability and innovatively decoupling the root causes of system instability into chemically and physically dominant failure pathways, this method achieves a shift from passive fault response to active trend prediction, from fuzzy fault diagnosis to precise failure attribution, and from universal intervention to differentiated intelligent regulation. Predictive proactive intervention is achieved through the stability index, identifying instability trends before quality deteriorates, and resolving the problem of delayed response.

[0044] 2. Generate differentiated intervention instructions based on diagnostic results to improve production efficiency, reduce waste and unnecessary downtime, and solve the problem of single intervention strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention, and those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0046] Figure 1 Flowchart of the method of the present invention.

[0047] Figure 2 It is a logic block diagram of the system of the present invention. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0049] Example 1:

[0050] See also Figure 1 The present invention provides a method for automatically detecting surface defects of printed products using machine vision, comprising the following steps: S1, synchronously collecting time-series defect image data, key component vibration data and substrate surface temperature data during the printing process to form a multimodal time-series data set; S2, extracting and quantifying key dynamic features from the multimodal time-series data set to generate a unified system state vector, wherein the key dynamic features include visual features, vibration features and thermodynamic features; S3, constructing a system state stability evolution model, inputting the system state vector into the stability evolution model, calculating and outputting a real-time system stability index, wherein the stability evolution model The model is constructed based on the deviation between the system state vector and the preset reference stable state vector; S4, in response to the real-time system stability index being lower than the preset warning threshold, the causal decoupling analysis is started, and the chemical dominant contribution and the physical dominant contribution are calculated according to the system state vector, and the current dominant failure path is determined by comparing the chemical dominant contribution and the physical dominant contribution; S5, based on the dominant failure path, differentiated intervention instructions are generated; if the dominant failure path is determined to be chemically dominated, an adjustment instruction for the ink mixing system is generated; if the dominant failure path is determined to be physically dominated, an early warning instruction for the equipment maintenance system is generated.

[0051] This embodiment provides a method for automatically detecting surface defects of printed products applied to a high-speed web-to-web gravure printing production line. In step S1, the method uses a high-precision clock source to synchronize the timestamps of machine vision cameras deployed at quality inspection stations, acceleration sensors installed on key bearing components such as the impression roller and the doctor blade support, and an infrared thermal imager array installed at the exit of a drying oven, thereby ensuring that the collected time-series defect image data, key component vibration data, and substrate surface temperature data are strictly aligned in time series to form a multimodal time series data set. The subsequent steps S2 to S5 are analyzed based on this unified data set, aiming to overcome the common problems of the prior art. The method overcomes the technical defects of delayed response and ambiguous causes; by constructing an evolutionary model that can prospectively evaluate system stability and creatively decoupling the root causes of system instability into chemical-dominated failure pathways and physical-dominated failure pathways, this method realizes the transformation from passive fault response to active trend prediction, from fuzzy fault diagnosis to precise failure attribution, and from universal intervention to differentiated intelligent regulation; its ultimate technical goal is to be able to identify the trend of system instability before printing defects occur on a large scale, and automatically generate adjustment or maintenance instructions with clear direction, so as to ensure product quality without interrupting production and significantly improve production efficiency and equipment health management level.

[0052] Example 2:

[0053] According to the aforementioned method, the S2 specifically includes: S21, based on the time-series defect image data, by calculating the image blur, texture entropy and color gradient norm, extracting and combining visual feature subvectors; S22, based on the vibration data of key components, by performing short-time Fourier transform on the data, extracting the energy amplitude at the relevant characteristic frequency of the key components, and combining them into vibration feature subvectors; S23, based on the substrate surface temperature data, by calculating the surface average temperature, temperature standard deviation and surface maximum temperature gradient, extracting and combining them into thermodynamic feature subvectors; S24, normalizing the visual feature subvector, the vibration feature subvector and the thermodynamic feature subvector and fusing them into the unified system state vector;

[0054] The core function of the system state vector construction in this embodiment is to transform the multi-source heterogeneous data extracted in steps S21 to S23 into a quantifiable, normalized and dimensionally unified system state vector through step S24. , providing an accurate numerical basis for subsequent stability modeling and causal decoupling; first, based on the time series defect image data , extract visual feature subvectors reflecting ink fluid properties and spreading state , which contains the image blur Characterize the leveling and texture entropy of ink Characterize the microscopic uniformity of the ink layer, color gradient norm The sharpness of the color edge is quantified. These three together constitute an indirect observation of chemical processes such as ink solvent volatilization; secondly, based on the vibration data of key components , through short-time Fourier transform analysis, extract the inherent characteristic frequency of the device The energy amplitude at , these amplitudes constitute the vibration characteristic subvectors that reflect the mechanical health status of the equipment ; Again, from the substrate surface temperature data Extract the thermodynamic characteristic subvectors that are closely related to the ink drying efficiency and energy transfer process , which contains the average surface temperature , temperature standard deviation and the maximum surface temperature gradient , which directly reflects the efficiency and uniformity of the drying process; finally, after normalization methods such as Z score normalization, these three feature sub-vectors are 、 and Fusion into a global high-dimensional system state vector: ;

[0055] Indicates time A vector describing the comprehensive state of the entire printing system;

[0056] express The visual feature sub-vector at the moment is composed of image blur, texture entropy, and color gradient norm;

[0057] express The vibration characteristic sub-vector at the moment is composed of the vibration amplitude at the characteristic frequency of the key component;

[0058] express The thermodynamic characteristic subvector at the moment is composed of the average surface temperature of the substrate, the temperature standard deviation, and the maximum surface temperature gradient;

[0059] Represents the transpose operation of a vector;

[0060] This vector at every moment It comprehensively and accurately describes the overall operating status of the entire printing system.

[0061] Example 3:

[0062] According to the aforementioned method, S3 specifically includes: S31, based on the system state vector and the preset reference stable state vector, calculating the difference between the two and obtaining a system state deviation vector; S32, performing a weighted quadratic calculation of a preset weight matrix on the system state deviation vector to generate a system potential energy value that characterizes the degree to which the system deviates from the stable point in the potential energy field; S33, based on the time series of the system potential energy value, calculating the product of its autocorrelation and variance and obtaining the inverse of the product to obtain the real-time system stability index;

[0063] According to the aforementioned method, the reference steady-state vector is obtained by continuously collecting and analyzing printing process data with a preset yield rate exceeding a specific standard, and calculating the time average of the calculated system state vector at each moment; the preset weight matrix is optimized by analyzing historical failure data using Lasso regression to quantitatively determine the contribution of each key dynamic feature to the final product quality defect;

[0064] The system state stability evolution modeling of this embodiment aims to achieve early warning of system critical transitions; its core is to introduce the system potential energy function and the real-time system stability index ;

[0065] System potential energy function: ;

[0066] Indicates that at the current moment The dimension is obtained by multimodal data fusion The real-time system state vector;

[0067] Represents a dimension The reference stable state vector represents the most ideal operating state of the system;

[0068] Indicates a The diagonal weight matrix of Representative The weight of the influence of each state feature on the system stability; represents the system state deviation vector, which is the calculation result of step S31;

[0069] Represents the potential energy value of the system, which is a scalar;

[0070] The design concept of this formula is derived from the theory of potential energy landscapes in physics. A healthy printing system is considered to be at the bottom of the potential energy field and in a stable state. When the system is subject to internal or external disturbances and begins to deviate from its optimal operating state, its potential energy value increases, just like a physical particle being pushed to a position with higher potential energy. This function aims to construct a mathematical model that can accurately quantify the degree to which the state of the entire high-dimensional system deviates from its ideal stable point using a single scalar, providing a clear and measurable basis for subsequent stability assessments.

[0071] In the real-time monitoring process, the system will collect and construct the real-time state vector Substitute this formula and continuously calculate the time series of potential energy values ;like If the value of continues to increase, it clearly indicates that the system is evolving from a stable state to an unstable critical state. This function condenses multi-dimensional and complex characteristic changes into a single indicator with clear physical meaning, making the judgment of the evolution trend of the overall state of the system intuitive and quantitative.

[0072] Select data from one or more production batches that have been confirmed to have a product quality rate higher than a specific standard (e.g., 99%); continuously collect and calculate the system state vector at each moment during the operation of these batches Finally, the arithmetic mean of all collected vectors is calculated over time, and the average vector is determined as ;

[0073] Collect a large amount of historical production data, including the system state vector at each moment and the corresponding final product defect rate; use the Lasso regression model for analysis, with each feature of the system state vector as the input variable and the product defect rate as the target variable; through the feature selection capability of Lasso regression, the contribution coefficient of each feature to the defect rate can be obtained, and these coefficients are normalized and used as the weight matrix The weight values on the corresponding diagonal lines , thus making the potential energy function more sensitive to those state changes that have historically been shown to be highly correlated with product defects;

[0074] Real-time system stability index : ;

[0075] represents the system potential energy value calculated by the above potential energy function;

[0076] It represents the autocorrelation coefficient value of the time series at the first-order lag, which is a dimensionless scalar;

[0077] It represents the variance of the time series in the same time window; step S33 is the complete calculation process of the index;

[0078] The theoretical basis of this index is the critical slowing-down phenomenon in complex system theory. When a dynamic system approaches its critical transition point, its speed of returning to equilibrium after a small disturbance will be significantly slowed down. This decline in recovery ability will be manifested as a simultaneous sharp increase in autocorrelation and variance in the time series of the system state variables. Only the potential energy value is monitored. The increase or decrease can only determine the deviation range, It aims to provide an earlier and more sensitive early warning signal of critical transition by capturing the dynamic characteristic of the system's slow recovery speed;

[0079] When the system is running, it calculates in real time based on a sliding time window The first-order autocorrelation coefficient and variance of the time series are obtained When the system is running stably, The fluctuation is small and the recovery is fast. Its autocorrelation and variance are both at a low level. The index maintains a high level of stable fluctuation; when the system begins to become unstable, the critical slowdown phenomenon leads to and The product of the two increases dramatically, resulting in the reciprocal of The index dropped sharply; therefore, by setting an early warning threshold and monitoring By checking whether the value falls below this value, high-sensitivity prediction can be achieved before large-scale quality problems occur.

[0080] Example 4:

[0081] According to the aforementioned method, the S4 specifically includes: S41, decoupling the system state vector and the preset weight matrix into a chemical part and a physical part, respectively, the chemical part corresponds to the visual and thermodynamic characteristics related to the ink, and the physical part corresponds to the vibration characteristics related to the equipment; S42, calculating the chemical dominant contribution based on the system state vector, the reference stable state vector and the weight matrix of the chemical part; S43, calculating the physical dominant contribution based on the system state vector, the reference stable state vector and the weight matrix of the physical part; S44, calculating the ratio of the chemical dominant contribution to the physical dominant contribution to obtain the dominant path ratio; S45, if the dominant path ratio is higher than the preset chemical dominant decision threshold, determining that the dominant failure path is chemically dominant; if the dominant path ratio is lower than the preset physical dominant decision threshold, determining that the dominant failure path is physically dominant;

[0082] According to the aforementioned method, the chemically dominant decision threshold and the physically dominant decision threshold are calibrated through controlled fault injection experiments, specifically including: introducing a perturbation of excessively rapid solvent evaporation into the ink system during the experiment, continuously recording and analyzing the statistical distribution of the dominant path ratio under this condition, thereby determining the chemically dominant decision threshold; and applying a known weak mechanical vibration excitation to a key rotating component during the experiment, continuously recording and analyzing the statistical distribution of the dominant path ratio under this condition, thereby determining the physically dominant decision threshold.

[0083] According to the aforementioned method, the preset warning threshold is determined by statistically analyzing a large amount of system data transitioning from normal state to fault state, and selecting a corresponding value that can maximize the warning recall rate while ensuring an acceptable false positive rate based on receiver operating characteristic curve analysis;

[0084] The causal decoupling analysis and decision-making technology of this embodiment aims to automatically and quantitatively diagnose the main driving factors leading to instability when the system issues an instability warning, thereby achieving precise intervention.

[0085] Causal contribution and pathway ratio: ; ; ;

[0086] represents the chemical state subvector composed of visual and thermodynamic features, which is the result of the decoupling of step S41;

[0087] Represents the physical state subvector composed of vibration characteristics, which is also the result of decoupling in step S41;

[0088] and Denote the reference steady-state vectors The corresponding chemical and physical subvectors in ;

[0089] and Represent the total weight matrix The submatrices in correspond to chemical and physical characteristics; Represents the norm of a vector, such as the L2 norm;

[0090] It represents the chemical dominant contribution, which is the calculation result of S42 and quantifies the contribution of the chemical factor deviation to the total potential energy gradient of the system;

[0091] It represents the physical dominant contribution, which is the calculation result of S43 and quantifies the contribution of the deviation of physical factors;

[0092] It represents the dominant pathway ratio, which is the calculation result of S44 and is a dimensionless scalar;

[0093] When the stability index When the warning threshold is breached, the system only knows that a failure is imminent, but cannot discern whether the root cause is chemical factors within the ink system or physical factors within the equipment's operation. To address this ambiguity, this embodiment proposes decomposing the total system potential energy gradient into two orthogonal components based on its physical and chemical origins: the chemical-dominant contribution and the physical-dominant contribution, thereby quantifying the strength of each component in driving system instability.

[0094] when When the alarm is triggered, the system immediately calculates in parallel and , and find the ratio of the two ; By By comparing with the preset decision threshold (S45), the system can automatically determine the root cause; a value much greater than 1 A value of 1 means that the contribution of the chemical part is significantly greater than that of the physical part, and the system is judged to be a chemical-dominated failure; conversely, a value far less than 1 This method directly solves the problem that traditional monitoring methods cannot distinguish whether the root cause of similar defect appearances is abnormal ink viscosity in the chemical field or mechanical vibration in the physical field.

[0095] By statistically analyzing a large amount of historical data from the transition phase from normal operation to failure, different The receiver operating characteristic curve under the threshold is the ROC curve; on this curve, select a point that can maximize the warning recall rate while ensuring an acceptable false positive rate. The value is determined as the warning threshold ;

[0096] Scientific calibration through controlled fault injection experiments; to determine the chemical dominant decision threshold In the experimental environment, a fast-evaporating solvent known to cause problems is artificially introduced into the ink system to actively induce chemical pathway failure and continuously record and analyze the state. The statistical distribution of , select the statistical lower bound of the distribution (such as the 5% quantile of the distribution) as Similarly, to determine the physical dominant decision threshold Apply a known weak mechanical vibration excitation to a key rotating component to simulate a physical path failure and record and analyze the The statistical distribution of ; This calibration method ensures the objectivity and robustness of the decision threshold.

[0097] Example 5:

[0098] According to the aforementioned method, S5 specifically includes: S51. If the dominant failure path is determined to be chemically dominant, automatically generating and sending instructions to the ink viscosity control system to adjust solvent replenishment parameters, and simultaneously pushing a prompt message indicating an abnormal ink status to the human-computer interaction interface; S52. If the dominant failure path is determined to be physically dominant, identifying the specific abnormally vibrating component based on the characteristic frequency information in the vibration characteristic subvector, and automatically generating a high-priority preventive maintenance work order for the component;

[0099] The generation of differentiated intervention instructions in this embodiment is the closed-loop execution link of the entire detection and diagnosis method, which directly reflects the accuracy of diagnosis and the efficiency of intervention. When the dominant failure path is determined, the precise differentiated intervention decision module of step S5 is activated. If the determination result of step S51 is chemical dominance, for example, Higher than , the module will immediately generate a control instruction, automatically increase the solvent replenishment pump frequency of the ink viscosity control system by a specific percentage through the industrial bus to deal with the identified excessive evaporation of the solvent, and at the same time push a clear prompt message to the operator's human-computer interaction interface, such as "The ink drying speed is abnormal, the system has automatically increased the solvent replenishment rate, please pay attention to subsequent quality changes"; if the judgment result of S52 is physical dominant, for example, it is detected Lower than , the module will further analyze the vibration characteristic sub-vector , according to the energy amplitude The largest characteristic frequency , identifies the specific source of abnormal vibration, such as the No. 3 guide roller bearing, from the mapping database of equipment components and characteristic frequencies, and automatically generates a high-priority preventive maintenance work order through the manufacturing execution system. The work order content clearly points to the faulty component, such as "Warning: The vibration of the No. 3 guide roller bearing exceeds the limit. It is recommended to arrange inspection or planned maintenance immediately." This intervention process is fully automatic and completed within seconds, realizing precise, closed-loop, and proactive intelligent control, effectively avoiding production stagnation and material waste caused by human misjudgment, intervention delays or improper operation.

[0100] Example 6:

[0101] An automatic detection system for printed surface defects using machine vision includes: a multimodal data synchronous acquisition module; a dynamic feature vector construction module; a system state stability evolution modeling module; a dominant failure path causal decoupling module; and a precise differentiated intervention decision module.

[0102] This embodiment also provides a system for automatically detecting surface defects of printed matter for implementing the above method; the system comprises the following functional modules:

[0103] The multimodal data synchronization acquisition module, whose hardware foundation includes a high-resolution linear array CCD camera, piezoelectric accelerometers installed at key locations on the equipment, an infrared thermal imager array, and a network time protocol server to ensure time consistency, is responsible for accurately and synchronously acquiring three data streams: visual, vibration, and temperature.

[0104] The dynamic feature vector building block is usually materialized as an edge computing unit or embedded system, which has internally solidified image processing algorithms (for calculating image blur, texture entropy and color gradient norm), digital signal processing algorithms (for performing short-time Fourier transform and extracting energy amplitude) and thermal imaging analysis algorithms to convert massive amounts of raw sensor data into a unified and normalized system state vector in real time. ;

[0105] The system state stability evolution modeling module is a software package deployed on an industrial PC or server. This software module continuously receives Vector flow, and based on the pre-calibrated reference steady-state vector and the weight matrix , real-time calculation of system potential energy and stability index , thereby achieving quantitative evaluation of system operation status and prediction of stability trend;

[0106] The leading failure path causal decoupling module is a core logic component in the aforementioned software package; when the system stability index is lower than the preset warning threshold This module is triggered by the condition, which is responsible for performing the decomposition calculation of the dominant contribution of chemistry and physics, and The comparison result with the decision threshold outputs a clear judgment on the root cause of instability;

[0107] The precise and differentiated intervention decision-making module serves as the final output and execution interface of the system. Based on the judgment conclusion of the decoupling module, this module sends control instructions directly to the programmable logic controller of the ink viscosity control system through the industrial field bus protocol, or pushes electronic preventive maintenance work orders containing precise fault location information to the management terminal of the equipment maintenance department through the enterprise resource planning system.

[0108] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for automatically detecting surface defects of printed matter using machine vision, characterized in that: The steps include: S1. Synchronously collect time-series defect image data, key component vibration data, and substrate surface temperature data during the printing process to form a multimodal time-series dataset; S2. Extracting and quantifying key dynamic features from the multimodal time series dataset to generate a unified system state vector; the key dynamic features include visual features, vibration features, and thermodynamic features; S3. Constructing a system state stability evolution model, inputting the system state vector into the stability evolution model, and calculating and outputting a real-time system stability index; the stability evolution model is constructed based on the deviation between the system state vector and a preset reference stable state vector; S4. In response to the real-time system stability index being lower than a preset warning threshold, initiating causal decoupling analysis, calculating a chemical dominant contribution and a physical dominant contribution based on the system state vector, and determining a current dominant failure path by comparing the chemical dominant contribution and the physical dominant contribution; S5. Generate differentiated intervention instructions based on the dominant failure path; if the dominant failure path is determined to be chemically dominant, generate adjustment instructions for the ink mixing system; If the dominant failure path is determined to be physically dominant, an early warning instruction is generated for the equipment maintenance system.

2. The method for automatically detecting surface defects of printed matter using machine vision according to claim 1, characterized in that: The S2 specifically includes: S21, based on the temporal defect image data, extracting and combining visual feature sub-vectors by calculating image blur, texture entropy, and color gradient norm; S22. Based on the vibration data of the key components, extract the energy amplitude at the relevant characteristic frequency of the key components by performing short-time Fourier transform on the vibration data of the key components, and combine them into a vibration characteristic subvector; S23, based on the substrate surface temperature data, by calculating the surface average temperature, temperature standard deviation and surface maximum temperature gradient, extracting and combining them into a thermodynamic characteristic subvector; S24 , normalizing the visual feature sub-vector, the vibration feature sub-vector, and the thermodynamic feature sub-vector and fusing them into the unified system state vector.

3. The method for automatically detecting surface defects of printed matter using machine vision according to claim 1, characterized in that: The S3 specifically includes: S31. Calculate the difference between the system state vector and the preset reference stable state vector to obtain a system state deviation vector; S32, performing a weighted quadratic calculation of a preset weight matrix on the system state deviation vector to generate a system potential energy value representing the degree to which the system deviates from a stable point in the potential energy field; S33. Based on the time series of the system potential energy value, calculate the product of the time series autocorrelation and variance of the system potential energy value, and obtain the inverse of the product to obtain the real-time system stability index.

4. The method for automatically detecting surface defects of printed matter using machine vision according to claim 3, characterized in that: The reference steady-state vector is obtained by continuously collecting and analyzing printing process data with a preset yield rate higher than a specific standard, and calculating the time average of the calculated system state vectors at each moment; The preset weight matrix is optimized by analyzing historical failure data using Lasso regression to quantitatively determine the contribution of each key dynamic feature to the final product quality defects.

5. The method for automatically detecting surface defects of printed matter using machine vision according to claim 4, characterized in that: The S4 specifically includes: S41, decoupling the system state vector and the preset weight matrix into a chemical part and a physical part, respectively, wherein the chemical part corresponds to the visual and thermodynamic characteristics related to the ink, and the physical part corresponds to the vibration characteristics related to the device; S42, calculating the chemical dominant contribution based on the system state vector, the reference stable state vector and the weight matrix of the chemical part; S43, calculating the physical dominant contribution based on the system state vector, the reference stable state vector and the weight matrix of the physical part; S44, calculating the ratio of the chemical dominant contribution to the physical dominant contribution to obtain a dominant pathway ratio; S45. If the dominant pathway ratio is higher than the preset chemical-dominant decision threshold, the dominant failure pathway is determined to be chemically dominant; if the dominant pathway ratio is lower than the preset physical-dominant decision threshold, the dominant failure pathway is determined to be physically dominant.

6. The method for automatically detecting surface defects of printed matter using machine vision according to claim 5, characterized in that: The chemical-dominant decision threshold and the physical-dominant decision threshold are calibrated through controlled fault injection experiments, specifically including: a) introducing a perturbation of excessively fast solvent evaporation into the ink system during the experiment, continuously recording and analyzing the statistical distribution of the dominant pathway ratio under this condition, so as to determine the chemical dominance decision threshold; b) By applying known weak mechanical vibration excitation to key rotating components in the experiment, the statistical distribution of the dominant path ratio under this condition is continuously recorded and analyzed to determine the physical dominant decision threshold.

7. The method for automatically detecting surface defects of printed matter using machine vision according to claim 1, characterized in that: The preset warning threshold is determined by statistically analyzing a large amount of system data transitioning from normal state to fault state, and based on receiver operating characteristic curve analysis, selecting a corresponding value that can maximize the warning recall rate while ensuring an acceptable false positive rate.

8. The method for automatically detecting surface defects of printed matter using machine vision according to claim 1, characterized in that: The S5 specifically includes: S51. If the dominant failure path is determined to be chemically dominant, a command is automatically generated and sent to the ink viscosity control system to adjust the solvent replenishment parameters, and a prompt message indicating an abnormal ink status is pushed to the human-computer interaction interface; S52: If the dominant failure path is determined to be physically dominant, then based on the characteristic frequency information in the vibration characteristic subvector, the specific abnormal vibration component is identified, and a high-priority preventive maintenance work order for the component is automatically generated.

9. A system for automatically detecting surface defects of printed matter using machine vision, applying the method for automatically detecting surface defects of printed matter using machine vision according to any one of claims 1 to 8, characterized in that: include: Multimodal data synchronous acquisition module, used to synchronously collect time-series defect image data, key component vibration data and substrate surface temperature data during the printing process to form a multimodal time-series data set; A dynamic feature vector construction module is used to extract and quantify key dynamic features reflecting the system state from the multimodal time series data set and fuse them into a unified system state vector; A system state stability evolution modeling module is used to construct a model describing the dynamic evolution of the system state vector based on the system state vector, and calculate and output a real-time system stability index based on the model; a dominant failure path causal decoupling module, configured to be activated in response to a condition where the real-time system stability index falls below a preset warning threshold to analyze the system state vector, thereby decoupling and determining the chemically dominant or physically dominant failure path that causes system instability; The precise differentiated intervention decision module is used to generate and output differentiated control instructions for the chemical pathway or the physical pathway according to the dominant failure pathway determined by the causal decoupling module.

Images