A multi-modal data fusion digital printing device color dynamic calibration system

Abstract

The application relates to the field of digital printing equipment, and discloses a multi-modal data fusion digital printing equipment color dynamic calibration system, which comprises a data acquisition and preprocessing module, a tensor modeling and optimization module, a causal reasoning module, a feedback control module, a cross-equipment mapping module and a self-diagnosis module; the method comprises the following steps: through real-time acquisition of multi-modal data of equipment and environment, intelligent analysis and dynamic adjustment of control parameters, the problem that traditional color management methods lack real-time feedback and dynamic adaptation capability is solved, and the stability and accuracy of the color calibration process are ensured. Through acquisition of color, image, environment and equipment operation data, construction of multi-modal tensors and sparse feature compression, combination of causal structure learning, realization of color deviation reasoning and adjustment amount generation, auxiliary self-diagnosis mechanism and closed-loop control, real-time monitoring of the equipment operation state and dynamic adjustment of the color state are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of digital printing equipment, in particular to a multi-modal data fusion digital printing equipment color dynamic calibration system. BACKGROUND

[0002] In the current color management process of digital printing equipment, the traditional scheme relies on static color calibration model and pre-established device operation parameter adjustment strategy. In the actual printing task continuous operation, affected by environmental conditions change and equipment state fluctuation and other factors, the color output of the equipment often produces dynamic deviation. Due to the lack of real-time perception and modeling ability of the environment and the equipment state in the running process of the traditional system, it is difficult to adjust the control parameters in time, and it is easy to appear the problems of poor color stability and low output consistency.

[0003] Some existing solutions try to introduce feedback adjustment mechanism or image monitoring based color calibration method, but there are the following technical limitations: on the one hand, the data collected by the system is limited, which cannot fully reflect the state change of the printing equipment in the running process, especially the lack of multi-modal data fusion ability; on the other hand, there is a lack of modeling mechanism based on causal relationship, the system often relies on experience rules or static model for adjustment, and it is difficult to realize accurate identification and dynamic reasoning of the color deviation causes.

[0004] In addition, the existing methods generally do not consider the unified structured modeling of environmental parameters, equipment operation parameters, image data and color data, which limits the expression ability and generalization ability of the model to some extent. At the same time, the traditional calibration system often relies on manual intervention to perform color correction process, lacks automation, and cannot meet the requirements of color stability and process robustness of current high frequency and high precision printing operation.

[0005] Therefore, the present application provides a multi-modal data fusion digital printing equipment color dynamic calibration system to solve the deficiencies of the prior art. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a multi-modal data fusion digital printing equipment color dynamic calibration system, which solves the problem of lack of real-time feedback and dynamic adaptation ability in the traditional color management method.

[0007] To achieve the above purpose, the present application is realized by the following technical scheme: a multi-modal data fusion digital printing equipment color dynamic calibration system, the system comprises the following modules:

[0008] A data acquisition and preprocessing module is configured to acquire multi-modal raw data from a digital printing device, to perform denoising processing, missing value completion, numerical normalization, and time alignment operations, and to generate multi-modal preprocessed data;

[0009] A tensor modeling and optimization module is configured to construct the multi-modal preprocessed data into a unified dimension tensor expression form, to implement tensor feature optimization based on the importance of each modal data, and to obtain a sparse feature structure;

[0010] A causal reasoning module is configured to construct a causal graph model of the causal relationship between variables contained in the sparse feature structure, to identify key factors affecting color deviation, and to output a reasoning result of the cause of the deviation;

[0011] A feedback control module is configured to combine the reasoning result with a current color state and a preset color target to generate corresponding device adjustment parameters for dynamically adjusting the operating state of the digital printing device;

[0012] A cross-device mapping module is configured to establish a modal structure mapping relationship between different types of digital printing devices, and to migrate the tensor expression structure and the causal model on an existing device to a target device, so that the feedback control module maintains the continuity of the calibration function in heterogeneous devices;

[0013] A self-diagnosis module is configured to acquire various operating state data and environmental variables during operation, to generate a device health evaluation result based on multi-dimensional indicators, and to trigger the feedback control module to perform parameter correction and control strategy update when an abnormal device operating state is detected.

[0014] Preferably, the data acquisition and preprocessing module includes the following steps:

[0015] Multi-modal raw data is acquired from a digital printing device, including color data, image data, environmental parameter data, and device operating parameter data;

[0016] Time alignment operations are performed on the multi-modal raw data based on a unified time reference benchmark;

[0017] Noise processing is performed on the time-aligned data, including filtering operations based on local statistical methods;

[0018] In the presence of missing data, numerical completion based on an interpolation function is used to complete missing value filling;

[0019] The processed multi-modal data is normalized according to a unified numerical range, and the normalization is a linear normalization mapping.

[0020] Preferably, the tensor modeling and optimization module includes the following steps:

[0021] receiving the pre-processed multi-modal data, constructing a tensor expression structure with unified dimensions according to the structural characteristics of each modal data in time, space and semantic dimensions;

[0022] After the tensor construction is completed, the influence degree of each modal in color deviation modeling is analyzed, and each modal is given a corresponding importance weight accordingly;

[0023] According to the modal weight, the tensor structure is processed for feature compression, and a sparse feature structure with strong representation ability and low data redundancy is extracted.

[0024] Preferably, the importance weight is determined based on an attention allocation mechanism, and the attention allocation mechanism dynamically adjusts the weight of each modal according to the response intensity and sensitivity of different modes in target prediction.

[0025] Preferably, the causal reasoning module comprises the following steps:

[0026] Receiving modal variables in the sparse feature structure as reasoning input, learning the causal structure according to the joint statistical characteristics between variables, and constructing a causal graph model containing multiple nodes and directed connections;

[0027] In the causal graph, each node represents a modal variable, and each edge represents the causal dependence relationship between two variables. The causal graph structure is determined by a data-driven structure search algorithm.

[0028] After the construction of the causal graph is completed, the target color deviation result is estimated by combining the observation variables, and the corresponding color deviation cause information is output through the conditional probability propagation mechanism.

[0029] Preferably, the feedback control module comprises the following steps:

[0030] Receiving the color deviation reasoning result provided by the causal reasoning module and the current color state, and comparing it with the preset target color state;

[0031] According to the difference between the current color state and the target color state, the adjustment amount is calculated for color correction;

[0032] The adjustment amount is input into the adjustment coefficient matrix for weighted adjustment to generate device control parameters;

[0033] The generated device control parameters are used to dynamically adjust the running state of the digital printing equipment, including inkjet control, ink flow rate, voltage signal and temperature parameters, for adjusting the output color of the equipment to the preset target value.

[0034] Preferably, the cross-device mapping module comprises the following steps:

[0035] receiving the tensor representation structure and the causal model in the source device, and extracting the modal structure information related to the target device running environment;

[0036] constructing a mapping function for inter-device conversion, and adjusting the tensor structure constructed in the source device in dimension and semantics to adapt to the modal characteristics of the target device;

[0037] According to the mapping function, the modeling parameters and control logic in the source device are converted into a structure suitable for the target device for continuous use after model migration.

[0038] Preferably, the cross-device mapping module further comprises the following steps:

[0039] After completing the model structure mapping, a small amount of sample data collected by the target device is used to adaptively adjust the migrated model;

[0040] The adaptive adjustment is based on an error feedback mechanism to dynamically correct the control parameters, so that the migrated model can meet the color calibration accuracy requirements on the target device and keep consistent with the running data of the target device.

[0041] Preferably, the self-diagnosis module comprises the following steps:

[0042] In the running process of the digital printing device, multiple running state data and environmental parameters including temperature, humidity, device load and ink viscosity are collected in real time;

[0043] According to a preset health assessment function, the collected data is calculated in multiple dimensions to generate a current health assessment result of the device;

[0044] The health assessment result is compared with a set threshold value, and if the assessment result is lower than the threshold value, it is judged that the device has potential abnormal risk, and the feedback control module is automatically triggered to dynamically correct the current device running parameters and control strategy, so as to ensure the continuous and stable color calibration process.

[0045] The application also provides a multi-modal data fusion digital printing device color dynamic calibration method, which comprises the following steps:

[0046] S1, obtaining multi-modal original data of a digital printing device, wherein the multi-modal original data comprises color data, image data, environmental parameter data and device running parameter data;

[0047] S2, performing time alignment, noise removal, missing completion and normalization processing on the multi-modal original data to generate preprocessed data with unified structure;

[0048] S3, constructing a multi-modal tensor structure based on the preprocessed data, and performing sparse feature compression after setting weights for each mode;

[0049] S4, based on sparse feature structure, causal structure learning is carried out, a causal graph model is constructed, and color deviation is inferred and estimated in combination with observation variables;

[0050] S5, comparing the inference result with the target color state, generating the adjustment amount required for color correction;

[0051] S6, adjusting the equipment operation parameters according to the adjustment amount, dynamically correcting the control logic of the digital printing equipment, and used for closed-loop control of the color state.

[0052] The application provides a multi-modal data fusion digital printing equipment color dynamic calibration system.

[0053] Has the following beneficial effects:

[0054] 1、The application realizes efficient fusion of multi-modal data under a unified structure by constructing a multi-modal tensor structure and performing sparse feature compression, so that various types of data have consistent expression in the model, and the accuracy of subsequent causal modeling and inference is improved. This processing method avoids modeling deviation caused by dimension mismatch or semantic inconsistency between different modal data, and enhances the analysis ability of the system to color state changes.

[0055] 2、The application introduces a causal structure learning method, not only captures the correlation between variables, but also establishes a causal graph model reflecting the control mechanism, so that the inference of color deviation is more explainable and targeted. The construction process of the causal model integrates observation variable distribution and conditional independence analysis, effectively supports the generation of the adjustment amount and the update of the control strategy, and meets the color dynamic regulation and control requirements under the influence of multiple factors.

[0056] 3、The application supports the migration and adaptation of the tensor structure and the causal model between different devices through the cross-device mapping module design, significantly enhancing the universality of the system in heterogeneous printing equipment. Through the dimension and semantic adjustment of the mapping function, the continuity and consistency of the color calibration model between devices with large differences in hardware configuration can be maintained, thereby reducing the resource consumption of repeated modeling.

[0057] 4、The application can monitor the equipment running state and environmental parameters in real time through the self-diagnosis module, quantitatively evaluate the health status of the equipment based on the health evaluation function, and automatically link the feedback control module to implement correction measures when detecting abnormalities. This mechanism ensures that the equipment can still maintain a stable color calibration process under non-ideal operating conditions, effectively reducing the time delay risk of fault intervention.

[0058] 5、The application can automatically adjust the control parameters according to the color deviation obtained by reasoning, realize dynamic correction and continuous calibration of the color state, has strong response ability in the continuous printing process, is suitable for processing color output fluctuation under complex working conditions and variable operating conditions, and improves the stability and reliability of the overall operation of the system. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 The system architecture of the application is shown in the figure.

[0060] Figure 2 The method flowchart of the application is shown in the figure. DETAILED DESCRIPTION

[0061] The technical solutions in the embodiments of the application will be described clearly and completely in combination with the drawings of the application specification. Obviously, the described embodiments are only part of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0062] Please refer to Figure 1 The embodiment of the application provides a multi-modal data fusion digital printing equipment color dynamic calibration system, which comprises the following modules:

[0063] The data acquisition and preprocessing module is used for acquiring multi-modal original data from the digital printing equipment, performing denoising, missing value completion, numerical normalization and time alignment operations, and generating multi-modal preprocessed data.

[0064] In the embodiment, the data acquisition and preprocessing module acquires multi-modal original data from the digital printing equipment in real time during system operation. The original data includes but is not limited to color data output by the equipment, image data generated during equipment operation, environmental parameter data such as temperature and humidity related to equipment operation environment, and equipment operation parameter data such as inkjet flow rate, voltage signal and work load. The above data modalities have characteristics such as large source difference, non-uniform scale, and inconsistent sampling frequency, so standardization operations are needed through preprocessing.

[0065] After data acquisition is completed, the module first performs time alignment processing on the above multi-modal data. The time alignment synchronously matches the data of different modalities by using a unified reference timestamp, so as to solve the deviation problem caused by data sampling delay and different frequencies. Preferably, an interpolation strategy or dynamic time warping method is used to preserve the original data characteristics while ensuring alignment accuracy. After alignment, all modal data will be reorganized into a data structure based on a unified time axis to support subsequent tensor modeling.

[0066] After time alignment, the system further processes the noise in the data. The noise processing method includes filtering operations based on local statistical characteristics, such as median filtering, weighted moving average, or wavelet-based denoising processing methods. The specific filtering strategy is dynamically selected according to the modal characteristics, for example, a small window median filter is preferred for color data to eliminate outliers, and an exponential moving average can be used for environmental parameter data to smooth the trend.

[0067] For the case of missing data in the data set, the system completes the missing values by interpolation function. Preferably, the interpolation method includes but is not limited to linear interpolation, spline interpolation, and interpolation method based on neighborhood weighting. In the modal data with missing points in time, the interpolation function takes the time neighborhood as the reference, combines the existing observation values to generate estimated values, and maintains the continuity of data changes. For the case of serious missing of some modalities, a regression model can also be constructed to assist in predicting missing values, thereby improving the accuracy of completion.

[0068] After noise processing and missing value completion, all modal data will be uniformly normalized. Normalization uses linear normalization mapping strategy to map the original data value to a unified numerical interval [0, 1] or [-1, 1], so as to eliminate the difference in order of magnitude and dimension of different modalities, and improve the comparability and processing efficiency of data in the subsequent modeling process. Assuming that the original data value is x i , the normalized result x i ' can be represented as:

[0069]

[0070] , x min represents the minimum value of the current modal data, x max represents the maximum value of the current modal data, and x i is any data point in the original data. This formula ensures that the normalized data has a unified scale, which helps data fusion and weight calculation in the process of multi-modal tensor modeling.

[0071] The normalized data will be stored in a structured and unified form and input into the subsequent tensor modeling and optimization module.

[0072] The tensor modeling and optimization module is used to construct the multi-modal preprocessed data into a unified dimension tensor expression form, and to implement tensor feature optimization based on the importance of each modal data, to obtain a sparse feature structure.

[0073] In this embodiment, the tensor modeling and optimization module receives the multi-modal data processed by the data acquisition and preprocessing module. The multi-modal data that has undergone time alignment, noise processing, missing value completion, and normalization already has a unified time scale and numerical range. On this basis, the module analyzes each modality data and constructs a unified dimensional tensor expression structure based on its structural characteristics in the time, space, and semantic dimensions. Specifically, each dimension of the tensor represents different data characteristics, such as time, space, or specific parameters of each modality. The construction of the tensor follows the following principles:

[0074] Time dimension: The data of all modalities will be aligned according to a unified timestamp to ensure the consistency of the tensor in time.

[0075] Spatial dimension: The spatial characteristics of each modality will be mapped to the same spatial scale, ensuring the spatial consistency of different modal data.

[0076] Semantic dimension: The semantic information contained in the data of different modalities, such as color data, image data, and environmental parameters, will be converted into the same representation to allow reasonable fusion in the tensor structure.

[0077] After completing the construction of the tensor structure, the module enters the stage of assigning the importance weight of each modality. The influence of each modality in color deviation modeling is different, so the module analyzes the response intensity and sensitivity of each modality data in target prediction to determine its importance in the tensor. Preferably, this process uses an attention allocation mechanism to dynamically adjust the weight of each modality. The attention allocation mechanism automatically assigns a corresponding weight to each modality based on the data characteristics of each modality and its contribution to the color prediction result. The core idea of this mechanism is to dynamically adjust the importance of each modality by calculating its performance on historical data and its impact on the current task, so that the model pays more attention to the modality data that has a greater impact on the prediction result. Specifically, the following attention weight calculation formula can be used:

[0078]

[0079] where w i represents the weight of the i-th modality; a i is the attention score calculated based on the performance of the modality in historical data, and j is the index of all modalities. Through this formula, the contribution of each modality can be converted into its weight value, thereby optimizing the information fusion in the tensor expression.

[0080] After the weight assignment is completed, the tensor modeling and optimization module further performs feature compression processing to extract sparse feature structures with low data redundancy and strong representation ability. The goal of feature compression is to reduce the dimensionality of the data and remove unnecessary redundant information, thereby improving computational efficiency and reducing model complexity. This process preferably uses sparse matrix decomposition, principal component analysis (PCA), or other constraint optimization-based feature selection methods. Through these methods, it is possible to eliminate irrelevant features that contribute little to the prediction result while retaining important features of the data. Specifically, the compression process can be represented by the following mathematical formula:

[0081] X' = Φ(X);

[0082] where X is the original tensor data, Φ represents the compression operation, and X' is the sparse feature tensor after compression. The compressed tensor will retain the most representative features and eliminate redundant features, thereby optimizing the training and inference process of the subsequent model.

[0083] The causal reasoning module constructs a causal graph model between the variables contained in the sparse feature structure, identifies key factors affecting color deviation, and outputs reasoning results of the causes of deviation.

[0084] In this embodiment, the causal reasoning module is used to identify and analyze key factors causing color deviation based on the sparse feature structure output by the tensor modeling and optimization module, forming a data-driven causal model with explanatory power, thereby providing mechanism-level support and guidance for the color dynamic calibration process.

[0085] The input of the causal reasoning module is the sparse feature tensor after feature compression and modal fusion processing. This sparse feature structure has fully retained the representative parameters of each key modality and eliminated redundant or noisy interference features. The module constructs a causal graph model based on these feature variables, and the core of this model is to identify the causal dependence relationship between variables, rather than just correlation modeling.

[0086] To achieve the above goal, the module first performs causal structure learning based on the joint probability distribution characteristics of each variable in the sparse feature structure. The structure learning uses a data-driven search strategy to determine the topology of the causal graph together with a scoring function and constraint conditions. Specifically, the module uses structure learning algorithms based on the score-search paradigm, such as Greedy-Equivalence-Search (GES) or neighborhood-restricted Bayesian network structure learning methods. The structure learning process can be abstracted as the following mathematical modeling problem:

[0087] Let the variable set be Its joint distribution is P(v1,v2,…,v n), the goal is to find a Directed-Acyclic-Graph (DAG) where each edge e ij represents the direct causal influence relationship between variables v i and v j . The structure learning process optimizes the graph structure by minimizing the following objective function:

[0088]

[0089] where, is the scoring function that measures the predictive power of each variable given its parent set Pa(v i ; G) under the current graph structure G. The scoring function is preferably the Bayesian Information Criterion (BIC) or Akaike Information Criterion (AIC).

[0090] In the causal graph model construction process, each node represents a modal variable in the sparse feature structure, such as color parameters, environmental humidity, inkjet pressure, etc.; each directed edge represents the direct causal dependence relationship between two variables. For example, if edge v i → v j exists, it means that variable v i is a direct dependent variable of variable v j . The causal graph model can not only be used for static causal relationship modeling, but also can be extended to a Temporal-Bayesian-Network to adapt to non-stationary characteristics when the system running environment has time dynamic changes.

[0091] When the causal graph is constructed, the module further combines the observation data to perform inference operations to estimate the causes of the target variable (i.e., the color deviation index). The inference process is completed based on the conditional probability propagation mechanism in the Bayesian network. Let the color deviation result be the target variable y, its prior distribution be P(y), and the observation variable set be O = {o1, o2, …, o k}, then the posterior inference target is:

[0092]

[0093] where the joint probability P(y, O) can be expanded according to the parent-child relationship of the nodes in the causal graph by the chain rule:

[0094]

[0095] In the reasoning process, the Variable-Elimination or Forward-Backward-Propagation algorithm is adopted, and the possible sources of the target deviation are quantitatively evaluated in combination with the current value of the observed variable, and the contribution degree of each candidate dependent variable to the color deviation is outputted.

[0096] To further enhance the explainability and practicality of the reasoning result, the module also introduces a causal path analysis mechanism to analyze the change path of the target variable. The path analysis is based on the path set from a specific input variable to the output variable y in the causal graph, and evaluates the cumulative influence degree of each path. The path influence degree is calculated as follows:

[0097]

[0098] wherein, δ p represents the cumulative influence degree of path p on y, θ ij is the edge weight corresponding to edge v i →v j The edge weight is learned from historical data and reflects the marginal influence of the dependent variable on the result variable.

[0099] Finally, the module outputs the causal reasoning result of the color deviation, which is presented in the form of a causal graph structure, an influence path, and a variable contribution degree ranking, providing a theoretical basis for subsequent color dynamic control strategies. The output result can also be used in the system-level feedback control mechanism to realize online correction of the color state.

[0100] The feedback control module combines the reasoning result with the current color state and the preset color target to generate corresponding device adjustment parameters for dynamically adjusting the running state of the digital printing equipment.

[0101] In this embodiment, the feedback control module is used to generate device control parameters for dynamically adjusting the running state of the digital printing equipment according to the difference between the color deviation reasoning result outputted by the causal reasoning module, the current color state, and the preset target color state, thereby realizing accurate correction of the color deviation.

[0102] In the workflow of the feedback control module, first, the color deviation reasoning result provided by the causal reasoning module is received, which contains the main causal factors affecting the color deviation and their corresponding contribution degrees. In addition, the module also acquires the color state of the current digital printing equipment, which is usually obtained through sensors and provides real-time feedback on the actual output color of the equipment. At this time, the current color state usually refers to the color characteristic parameters outputted by the printing equipment, such as chroma, saturation, brightness, etc. These parameters can accurately reflect the color characteristics exhibited by the equipment in actual work.

[0103] Next, the feedback control module compares the current color state with a preset target color state. The target color state refers to the desired ideal color state according to user requirements or device settings. This state is generally defined by a set color standard (such as the international color standard). The module determines the color amount that needs to be adjusted by calculating the difference between the current color state and the target color state.

[0104] After calculating the color difference, the module further calculates the adjustment amount based on this difference. This adjustment amount reflects the correction operation needed to achieve the target color state. Specifically, the calculation of the adjustment amount can be represented by the following formula:

[0105] ΔC = C target -C current ;

[0106] Where ΔC is the required adjustment amount, C target is the target color state, and C current is the current color state. This difference provides the feedback control module with the direction and magnitude of the adjustment needed.

[0107] After the adjustment amount is calculated, the module inputs this adjustment amount into a weighting adjustment coefficient matrix. The adjustment coefficient matrix is constructed based on the influence of each color adjustment parameter on the output color, usually involving multiple device parameters that affect color, such as inkjet control, ink flow rate, voltage signal, and temperature. These device parameters have complex causal relationships with the final output color, and different device parameters may have different contributions to color adjustment.

[0108] To ensure that each device parameter can be accurately adjusted according to the deviation of the current color state, the adjustment coefficient matrix assigns different weight values to each adjustment parameter. Specifically, the calculation process of the matrix can be expressed in the following form:

[0109] P = W·ΔC;

[0110] Where P is the device control parameter vector, W is the adjustment coefficient matrix, and ΔC is the required adjustment amount. The result of matrix multiplication is the final device control parameter. Through this process, the module can assign appropriate adjustment magnitudes to each device adjustment parameter according to the importance of different device adjustment parameters.

[0111] Finally, the generated device control parameters will be sent to the control system of the digital printing device for dynamic adjustment of the running state of the device. These adjustments include inkjet control (such as adjustment of ink drop size and jet frequency), ink flow rate, voltage signal, temperature parameter, etc. Through precise adjustment of these parameters, the color output of the device will gradually approach the preset target color state, thereby realizing the correction of color deviation.

[0112] The feedback control module can calibrate and optimize in real time during system operation according to the inference result and the current device state, thereby ensuring the consistency and stability of color output. In addition, since the process is based on causal inference and data-driven approach, it can effectively identify the main causes of color deviation and provide more scientific and reasonable decision basis for color adjustment.

[0113] The cross-device mapping module is used to establish a modal structure mapping relationship between different types of digital printing devices, and to migrate the tensor expression structure and causal model on the existing device to the target device, so that the feedback control module maintains the continuity of the calibration function in heterogeneous devices;

[0114] In this embodiment, the cross-device mapping module aims to migrate the tensor expression structure and causal model on the source device to the target device, to ensure that the color dynamic calibration function is continuously and consistently implemented between different types of digital printing devices.

[0115] The workflow of the cross-device mapping module first receives the tensor expression structure and causal model from the source device. The tensor structure of the source device is obtained through the previous model training process, which contains the multi-modal data features generated by the device in the color calibration process. The causal model reflects the causal dependence relationship between each device parameter and the color output. These information is migrated and adapted in the target device, so that the target device can inherit and perform the same calibration task.

[0116] The module first extracts the modal structure information related to the target device according to the running environment of the target device. The modal structure of the target device may have some differences with the source device, including differences in device hardware, changes in operating parameters, or specific modal variables related to the device working environment (such as environmental temperature, humidity, etc.). These modal structure information will be used to guide the following mapping process, to ensure that the migrated model can adapt to the characteristics of the target device.

[0117] After extracting the modal information related to the target device, the cross-device mapping module constructs a mapping function. The function's role is to adjust the tensor structure in the source device in terms of dimensions and semantics, so that it can adapt to the modal characteristics of the target device. Specifically, the mapping function not only considers the conversion of data dimensions, but also needs to handle the semantic differences specific to the device, such as the target device may use different parameter units, data range or color space. Therefore, the mapping function will transform the tensor structure in the source device so that it can be used on the target device.

[0118] The specific expression of the mapping function can be formalized as follows:

[0119]

[0120] where T target represents the tensor structure on the target device, T source is the tensor structure in the source device, is the modal characteristics of the target device, and f(·) is the mapping function. The output of the function is the adjusted target device tensor structure, ensuring that the target device can effectively calibrate the color based on the experience data of the source device.

[0121] The application of the mapping function is not limited to the adjustment of data structure, but also needs to convert the modeling parameters and control logic in the source device to adapt to the control system of the target device. These modeling parameters usually include device control parameters that affect color output, such as inkjet control, ink flow rate, voltage adjustment, etc. Control logic is the control strategy derived from the causal model on the source device, which includes how to adjust the device state according to the color deviation. Through the application of the mapping function, these parameters and logic of the source device will be converted into a form suitable for the target device.

[0122] After completing the mapping between devices, the cross-device mapping module further utilizes a small amount of sample data collected by the target device to adaptively adjust the migrated model. This process is based on an error feedback mechanism, which dynamically corrects the control parameters by calibrating the actual output of the target device, so that the migrated model can meet the color calibration accuracy requirements of the target device and keep consistent with the running data of the target device. Specifically, the target device collects actual running data (such as output color characteristics such as chroma, brightness, etc.), compares it with the target color state, and calculates the error between the current model and the target state. Then, the module will adjust the control parameters according to the error to optimize the color output of the device.

[0123] In the model adjustment process, the feedback mechanism can be quantitatively represented by the following formula:

[0124] ΔP adjust = g(ε);

[0125] where ΔP adjust represents the adjustment amount of the control parameter, ε is the current error, and g(·) is an error-based adjustment function. This adjustment function dynamically adjusts the control parameter according to the size of the error, so that the difference between the output color of the target device and the preset target color state is minimized.

[0126] Finally, the cross-device mapping module realizes model migration and adaptation between the source device and the target device through the above steps. The model after mapping and adaptive adjustment can continuously perform color calibration tasks on the target device, ensuring the consistency and accuracy of color output, and maintaining high continuity between different devices.

[0127] The self-diagnosis module is configured to collect various types of running state data and environmental variables during the operation of the digital printing device, evaluate the current running health of the device through multi-index fusion calculation, and automatically link the feedback control module for parameter correction and control strategy update when potential abnormalities are detected, thereby ensuring the continuous stability of the color calibration process.

[0128] In this embodiment, the self-diagnosis module is configured to collect multi-dimensional running state data and environmental variables in real time during the operation of the digital printing device, evaluate the current running health of the device through multi-index fusion calculation, and automatically link the feedback control module for parameter correction and control strategy update when potential abnormalities are detected, thereby ensuring the continuous stability of the color calibration process.

[0129] During system operation, the self-diagnosis module first collects multiple types of running state data and external environmental parameters from the device side. The collected data preferably includes but is not limited to device internal temperature, environmental humidity, device running load, ink viscosity, inkjet head working state, power supply voltage stability, system running frequency, and device usage time, etc. The above data is obtained in real time by multi-modal sensing units configured inside or outside the device, and is input into the subsequent evaluation calculation process in a unified data format.

[0130] To realize quantitative evaluation of the device running state, the self-diagnosis module performs fusion analysis on the collected multiple types of parameters based on a preset health evaluation function. This evaluation function considers the weight of each type of parameter in the device health evaluation and its historical evolution characteristics, and obtains the running health score of the current device by constructing a multi-dimensional index system. The basic expression of the evaluation function is as follows:

[0131]

[0132] where H represents the health evaluation score of the device, w i is the weight factor of the i-th state variable, and satisfies ∑w i = 1, x i represents the current collection value of the i-th state parameter, and f i(x i ) is a conversion function that maps the parameter value to a normalized health score. The function form can be linear, non-linear, or empirical rule-based, depending on the operational characteristics exhibited by the parameter in the device.

[0133] The weight factor w of each type of state parameter i is preferably set according to historical data statistics, empirical knowledge base, or machine learning algorithm training results, so that the overall evaluation result has dynamic adaptability. In order to further enhance the accuracy of the evaluation, historical trend comparison, time series sliding window processing and anomaly detection algorithm can also be introduced in the evaluation process to improve the response ability to sudden states.

[0134] After the evaluation is completed, the module compares the current health evaluation score H calculated with the set threshold H th . If the following conditions exist:

[0135] H < H th ;

[0136] The system determines that there is a potential risk of abnormal operation of the current device, and needs to execute the response mechanism. The threshold H th can be dynamically set according to the device model, usage frequency and operation strategy, and has a certain range of flexible adjustment.

[0137] When it is determined that the device has an abnormal risk, the self-diagnosis module will automatically generate an abnormality mark, and the feedback control module will start the adaptive correction process. The correction process includes real-time adjustment of device operating parameters and updating of control strategies. Specifically, the module calculates the adjustment amount of the control parameter to be corrected according to the difference between the current state parameter and the target health state. The correction process can be expressed in the following form:

[0138] ΔP = φ(H th -H);

[0139] Where ΔP is the correction amount of the control parameter, and φ(·) represents a control parameter mapping function based on the health evaluation difference. The function maps the change of health state to the change of device control layer instruction, such as inkjet pressure adjustment, ink heating strategy change or cooling unit activation, etc.

[0140] The correction process will directly act on the parameter system relied on by the feedback control module, and affect the input distribution of its control decision model. Through dynamic adjustment, the device can respond to the current potential operation abnormality in time, prevent the device state from further deterioration, and effectively avoid color deviation accumulation and output quality decline.

[0141] In addition, to enhance the stability of the system in long-term operation, the self-diagnosis module supports periodic health assessment and data recording functions, historical health status data can be stored and input into the data-driven model for continuous training and updating of the identification logic of the device state, so as to realize trend monitoring and predictive maintenance of the device operating condition in the long-term dimension.

[0142] Referring to Figure 2 The application also provides a multi-modal data fusion digital printing equipment color dynamic calibration method, which comprises the following steps:

[0143] S1, acquiring multi-modal original data of the digital printing equipment, the multi-modal original data comprising color data, image data, environmental parameter data and equipment operating parameter data;

[0144] The multi-modal original data covers color data, image data, environmental parameter data and equipment operating parameter data. The color data is usually collected by a color sensor built-in or externally connected to the equipment in real time, for reflecting the color performance of the actual printing output; the image data is collected by a high-speed camera device to acquire the structural information and spatial distribution of the equipment output image, for assisting in analyzing the stability of the color in different areas; the environmental parameter data includes but is not limited to temperature, humidity, air quality and other external variables affecting the color performance, which are collected by environmental sensors distributed around the equipment; and the equipment operating parameter data refers to the system operating state information recorded by the equipment during the working process, such as inkjet frequency, voltage, current, heating unit power and the like. These data are uniformly collected by a distributed collection system and provide a basis for subsequent processing.

[0145] S2, performing time alignment, noise removal, missing value completion and normalization processing on the multi-modal original data to generate preprocessed data with unified structure;

[0146] Firstly, the time stamp alignment operation is performed to realize the alignment of different modal data in the same time dimension, so that they have a time sequence corresponding relationship; secondly, the noise removal is performed based on the statistical method and the filtering algorithm to filter out abnormal values introduced due to equipment vibration, electromagnetic interference or collection error; then, the missing values in the data are completed by using the adjacent interpolation, model inference and other methods, so as to avoid the influence of incomplete sample information on the modeling process; finally, the numerical scale of various data is unified by normalization processing, so as to ensure that the data of different physical quantity units can be used in the same structure, which is convenient for subsequent tensor structure construction and feature analysis.

[0147] S3, constructing a multi-modal tensor structure based on the preprocessed data, and performing sparse feature compression after setting weights for each mode;

[0148] Tensor structure is a mathematical structure suitable for high-dimensional multi-modal data representation, which can represent the internal relationship of data in multiple modal dimensions. In the construction process, the corresponding weight coefficient is set for different modalities to reflect their relative importance in the overall modeling. Then, the system performs sparse feature compression operation on the constructed tensor structure, and retains key features by principal component extraction, tensor decomposition or dictionary learning, while eliminating redundant information and irrelevant variables, reducing model complexity and improving the interpretability of subsequent causal modeling.

[0149] S4, based on the sparse feature structure, learning the causal structure, constructing the causal graph model, and combining the observed variables to infer and estimate the color deviation;

[0150] In summary, the causal graph model is constructed by mining the causal dependence relationship between different modal features. The causal graph expresses the direct or indirect influence path between variables in the form of graph theory, so that the system can not only identify the correlation between variables, but also understand the mechanism of action. In the model construction process, the system introduces joint distribution analysis and conditional independence test strategy based on observed variables to identify key paths and action directions, and finally forms a causal graph structure for color deviation inference. This structure provides a basis for subsequent color state prediction and deviation source analysis.

[0151] S5, comparing the inference result with the target color state to generate the adjustment amount required for color correction;

[0152] In the inference process, the system uses the established causal path to calculate the deviation between the current color output and the expected state in combination with the current device state and collected data. After conversion and mapping, this deviation information forms a parameter adjustment amount that can be directly used for control layer, which covers the fine-tuning suggestions for multiple control variables such as inkjet amount and heating temperature.

[0153] S6, adjusting the device operating parameters according to the adjustment amount to dynamically correct the control logic of the digital printing device for closed-loop control of the color state;

[0154] Through feedback closed-loop control, the inference result is converted into specific parameter setting commands and applied to the control unit related to color calibration. The control logic continuously responds to the adjustment information input from the outside during operation, maintains the consistency between the color output and the target state, realizes the dynamic closed-loop regulation and control of the color state, and guarantees the stability of the printing quality under long-time operation.

[0155] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A color dynamic calibration system for digital printing equipment using multimodal data fusion, characterized in that, The system includes the following modules: The data acquisition and preprocessing module is used to acquire multimodal raw data from digital printing equipment to perform noise reduction, missing value completion, numerical normalization and time alignment operations to generate multimodal preprocessed data. The tensor modeling and optimization module is used to construct the multimodal preprocessed data into a tensor representation with a unified dimension, and to perform tensor feature optimization based on the importance of each modality to obtain a sparse feature structure. The tensor modeling and optimization module includes the following steps: receiving preprocessed multimodal data; constructing a tensor representation structure with a unified dimension based on the structural characteristics of each modality in the temporal, spatial, and semantic dimensions; after the tensor construction is completed, analyzing the influence of each modality on color deviation modeling, and assigning each modality a corresponding importance weight accordingly; and performing feature compression processing on the tensor structure based on the modality weights to extract a sparse feature structure with strong representational power and low data redundancy. The causal reasoning module constructs a causal graph model of the causal relationships between the variables contained in the sparse feature structure, identifies the key factors affecting color deviation, and outputs the reasoning results of the deviation causes. The feedback control module combines the inference results with the current color state and a preset color target to generate corresponding device adjustment parameters for dynamically adjusting the operating state of the digital printing equipment. The feedback control module includes the following steps: receiving color deviation inference results and the current color state provided by the causal inference module, and comparing them with a preset target color state; calculating an adjustment amount for color correction based on the difference between the current color state and the target color state; inputting the adjustment amount into an adjustment coefficient matrix for weighted adjustment to generate device control parameters; and using the generated device control parameters to dynamically adjust the operating state of the digital printing equipment, including adjustments to inkjet control, ink flow rate, voltage signal, and temperature parameters, to adjust the output color of the equipment to the preset target value. The cross-device mapping module is used to establish modal structure mapping relationships between different types of digital printing devices and to transfer the tensor expression structure and causal model on the existing devices to the target devices, so that the feedback control module can maintain the continuity of calibration function in heterogeneous devices. The self-diagnosis module is used to collect various operating status data and environmental variables during operation, generate equipment health assessment results based on multi-dimensional indicators, and trigger the feedback control module to perform parameter correction and control strategy update when an abnormal operating status of the equipment is detected.

2. The color dynamic calibration system for digital printing equipment based on multimodal data fusion according to claim 1, characterized in that, The data acquisition and preprocessing module includes the following steps: Multimodal raw data is acquired from digital printing equipment, including color data, image data, environmental parameter data, and equipment operating parameter data; Time alignment is performed on the multimodal raw data based on a unified time reference benchmark; The time-aligned data is subjected to noise processing, which includes filtering operations based on local statistical methods. In the case of missing data, a numerical completion method based on interpolation functions is used to fill in the missing values; The processed multimodal data is normalized according to a unified numerical range, and the normalization is a linear normalization mapping.

3. The color dynamic calibration system for digital printing equipment based on multimodal data fusion according to claim 1, characterized in that, The importance weights are determined based on an attention allocation mechanism, which dynamically adjusts the weights of each modality according to the response strength and sensitivity of different modalities in target prediction.

4. The color dynamic calibration system for digital printing equipment based on multimodal data fusion according to claim 1, characterized in that, The causal reasoning module includes the following steps: It receives modal variables from sparse feature structures as inference inputs, performs causal structure learning based on the joint statistical properties between variables, and constructs a causal graph model containing multiple nodes and directed connections. In the causal graph, each node represents a modal variable, and each edge represents a causal dependency between two variables. The structure of the causal graph is determined by a data-driven structure search algorithm. After the causal graph is constructed, the target color deviation result is estimated by combining the observed variables, and the corresponding color deviation cause information is output through the conditional probability propagation mechanism.

5. The color dynamic calibration system for digital printing equipment based on multimodal data fusion according to claim 1, characterized in that, The cross-device mapping module includes the following steps: Receive tensor representation structures and causal models from source devices, and extract modal structure information related to the operating environment of target devices; Construct a mapping function for inter-device conversion, which adjusts the tensor structure constructed in the source device in terms of dimension and semantics to adapt to the modal features of the target device; The modeling parameters and control logic in the source device are converted into a structural representation suitable for the target device based on the mapping function, for continuous use after model migration.

6. The color dynamic calibration system for digital printing equipment based on multimodal data fusion according to claim 5, characterized in that, The cross-device mapping module also includes the following steps: After completing the model structure mapping, the transferred model is adaptively adjusted using a small amount of sample data collected from the target device. The adaptive adjustment dynamically corrects the control parameters based on the error feedback mechanism, so that the transferred model can meet the color calibration accuracy requirements on the target device and remain consistent with the operating data of the target device.

7. The color dynamic calibration system for digital printing equipment based on multimodal data fusion according to claim 1, characterized in that, The self-diagnosis module includes the following steps: During the operation of digital printing equipment, various operating status data and environmental parameters, including temperature, humidity, equipment load, and ink viscosity, are collected in real time. Based on a preset health assessment function, the collected data is calculated in multiple dimensions to generate the current health assessment result of the device; The health assessment results are compared with a set threshold. If the assessment result is lower than the threshold, it is determined that the device has a potential abnormal risk. The feedback control module is automatically triggered to dynamically correct the current device operating parameters and control strategy to ensure the color calibration process remains stable.

8. A method for dynamic color calibration of digital printing equipment using multimodal data fusion, applied to the system described in any one of claims 1-7, characterized in that, The method includes the following steps: S1. Acquire multimodal raw data of digital printing equipment, wherein the multimodal raw data includes color data, image data, environmental parameter data and equipment operating parameter data; S2. Perform time alignment, noise removal, missing data completion, and normalization on the multimodal raw data to generate preprocessed data with a uniform structure. S3. Construct a multimodal tensor structure based on the preprocessed data, and perform sparse feature compression after setting weights for each mode; S4. Based on sparse feature structure, perform causal structure learning, construct a causal graph model, and combine observed variables to infer and estimate color deviation; S5. Compare the inference results with the target color state to generate the adjustment amount required for color correction; S6. Adjust the equipment operating parameters according to the adjustment amount, dynamically correct the control logic of the digital printing equipment, and use it for closed-loop control of color status.

Images