Multi-color laser marking intelligent regulation and control system based on multi-dimensional dynamic calibration

Abstract

The invention relates to the field of intelligent regulation and control, and particularly discloses a multi-color laser marking intelligent regulation and control system based on multi-dimensional dynamic calibration. The method comprises the following steps: firstly, acquiring laser marking environment temperature values at a plurality of preset time points acquired by a temperature sensor, laser equipment text parameters acquired by a database and laser marking material attribute text parameters, then performing feature extraction and correlation analysis on the three parameters by using a deep learning technology, and finally obtaining a classification result through a classifier. Therefore, the intelligent level of the laser marking process is improved, the production efficiency is further improved, the manual intervention and debugging difficulty are reduced, the production cost is reduced, and the reliability and high efficiency of the production process are ensured.

Description

Technical Field

[0001] The present application relates to the field of intelligent control, and more specifically, to a multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration. Background Art

[0002] Laser marking technology, a key branch of laser processing, is an advanced marking method that integrates laser, optics, precision mechanics, electronics, and computer technologies. It uses a computer-controlled laser beam as a processing method. Its basic principle is to use a computer-controlled, high-energy-density, focused laser beam to locally irradiate a metal workpiece along a predetermined trajectory, causing the surface material to instantly vaporize or change color. This allows text, patterns, barcodes, and other materials to be etched with a certain depth or color, leaving a permanent mark on the metal workpiece surface.

[0003] At present, although traditional marking machines can perform color marking, their energy efficiency, accuracy and adaptability are poor, and they cannot meet modern high-precision and diversified needs.

[0004] Therefore, an intelligent control system for multi-color laser marking based on multi-dimensional dynamic calibration is desired. Summary of the Invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration. The system first obtains the laser marking environment temperature values ​​at multiple predetermined time points collected by a temperature sensor, the laser equipment text parameters collected by a database, and the laser marking material attribute text parameters. Then, using deep learning technology, the system performs feature extraction and correlation analysis on the three. Finally, a classification result is obtained through a classifier to determine whether the multi-color laser marking equipment parameters need to be adjusted, thereby improving the intelligence level of the laser marking process, thereby improving production efficiency, reducing manual intervention and debugging difficulty, reducing production costs, and ensuring the reliability and efficiency of the production process.

[0006] According to one aspect of the present application, a multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration is provided, which includes:

[0007] A multi-color laser marking data acquisition module is used to obtain the laser marking environment temperature values ​​at multiple predetermined time points collected by the temperature sensor, the laser equipment text parameters collected by the database, and the laser marking material property text parameters;

[0008] a multi-color laser marking data extraction module, configured to extract a laser marking ambient temperature feature vector and a laser marking parameter multimodal semantic feature vector from the laser marking ambient temperature values ​​at a plurality of predetermined time points acquired by the temperature sensor, the laser equipment text parameters acquired by the database, and the laser marking material attribute text parameters;

[0009] The multi-color laser marking equipment parameter adjustment judgment module is used to judge whether the multi-color laser marking equipment parameters need to be adjusted based on the laser marking environment temperature feature vector and the laser marking parameter multimodal semantic feature vector.

[0010] Compared with the existing technology, the present application provides a multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration, which first obtains the laser marking ambient temperature values ​​at multiple predetermined time points collected by the temperature sensor, the laser equipment text parameters and the laser marking material property text parameters collected by the database, and then uses deep learning technology to perform feature extraction and correlation analysis on the three. Finally, the classification results are obtained through the classifier to determine whether the parameters of the multi-color laser marking equipment need to be adjusted, thereby improving the intelligence level of the laser marking process, and then improving production efficiency, reducing manual intervention and debugging difficulty, reducing production costs, and ensuring the reliability and efficiency of the production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0012] Figure 1 4 is a block diagram of a multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to an embodiment of the present application.

[0013] Figure 2 4 is a block diagram of a multi-color laser marking data extraction module in a multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to an embodiment of the present application.

[0014] Figure 3 This is a block diagram of a parameter adjustment and judgment module for a multi-color laser marking device in a multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to an embodiment of the present application.

[0015] Figure 4 Schematic diagram of the overall architecture of a multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to an embodiment of the present application.

[0016] Figure 5 Schematic diagram of grating dynamic adjustment of a multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to an embodiment of the present application.

[0017] Figure 6 This is a flow chart of the intelligent control module of the multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0019] Figure 1 FIG is a block diagram of a multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to an embodiment of the present application. Figure 1 As shown, according to an embodiment of the present application, the multi-color laser marking intelligent control system 100 based on multi-dimensional dynamic calibration includes: a multi-color laser marking data acquisition module 110, which is used to obtain the laser marking ambient temperature values ​​at multiple predetermined time points collected by the temperature sensor, the laser equipment text parameters and the laser marking material attribute text parameters collected by the database; a multi-color laser marking data extraction module 120, which is used to extract the laser marking ambient temperature feature vector and the laser marking parameter multimodal semantic feature vector from the laser marking ambient temperature values ​​at multiple predetermined time points collected by the temperature sensor, the laser equipment text parameters collected by the database and the laser marking material attribute text parameters; a multi-color laser marking device parameter adjustment judgment module 130, which is used to judge whether the multi-color laser marking device parameters need to be adjusted based on the laser marking ambient temperature feature vector and the laser marking parameter multimodal semantic feature vector.

[0020] In the above-mentioned multi-color laser marking intelligent control system 100 based on multi-dimensional dynamic calibration, the multi-color laser marking data acquisition module 110 is used to obtain the laser marking environment temperature values ​​at multiple predetermined time points collected by the temperature sensor, the laser equipment text parameters collected by the database, and the laser marking material property text parameters. It should be understood that in the technical solution of the present application, through data integration and analysis, it is possible to effectively track and optimize the impact of temperature and equipment status on the marking quality during the laser marking process. Specifically, laser marking technology is a key branch in the field of laser processing, combining a variety of advanced technologies such as laser technology, optical principles, precision machinery, electronics and computer technology. Its working principle is to control the laser beam through a computer, and irradiate the high-energy-density focused laser onto the surface of the metal workpiece according to a set trajectory, so that the local material quickly evaporates or undergoes physical and chemical changes, changes its color or produces an etching effect, thereby forming a permanent mark on the surface of the workpiece, such as text, patterns and barcodes. At present, metal laser color marking usually adopts YAG solid laser, which has the characteristics of short wavelength and high energy density, so that the laser energy it produces can be effectively absorbed by most metal materials, and can be focused into a smaller spot, which is suitable for achieving high-precision metal marking. However, YAG lasers also have a large heating problem, which may affect the stability and service life of the equipment. In addition, the existing metal laser color marking process involves the coordination of multiple parameters and requires precise adjustment, so it is very difficult to debug and the processing efficiency is low. Therefore, in the technical solution of the present application, by obtaining the laser marking ambient temperature values ​​at multiple predetermined time points collected by the temperature sensor, the laser equipment text parameters and the laser marking material attribute text parameters collected by the database, and combining deep learning technology, it is determined whether the parameters of the multi-color laser marking equipment need to be adjusted, so as to improve the intelligence level of the laser marking process, thereby optimizing production efficiency, reducing manual intervention and debugging difficulty, reducing production costs, and ensuring the stability and efficiency of the production process.

[0021] In the aforementioned multi-color laser marking intelligent control system 100 based on multi-dimensional dynamic calibration, the multi-color laser marking data extraction module 120 is configured to extract a laser marking environment temperature feature vector and a laser marking parameter multimodal semantic feature vector from the laser marking environment temperature values ​​at multiple predetermined time points collected by the temperature sensor, the laser equipment parameters collected by the database, and the laser marking material attribute parameters. It should be understood that by fusing the environment temperature feature vector, the equipment parameter feature vector, and the material attribute feature vector, a multimodal semantic feature vector can be constructed. This feature vector integrates multidimensional information such as temperature, equipment status, and material properties, and can comprehensively describe the interactions between various factors in the laser marking process.

[0022] Figure 2FIG. 1 is a block diagram of a multi-color laser marking data extraction module in a multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to an embodiment of the present application. Figure 2 As shown, in a specific embodiment of the present application, the multi-color laser marking data extraction module 120 includes: a laser marking environment temperature value feature extraction unit 121, which is used to perform feature extraction on the laser marking environment temperature values ​​at multiple predetermined time points collected by the temperature sensor to obtain the laser marking environment temperature feature vector; a laser equipment text parameter feature extraction unit 122, which is used to perform feature extraction on the laser equipment text parameters collected by the database to obtain a laser equipment text semantic feature vector; a laser marking material attribute text parameter feature extraction unit 123, which is used to perform feature extraction on the laser marking material attribute text parameters to obtain a laser marking material attribute text semantic feature vector; a laser marking parameter multimodal feature association unit 124, which is used to associate the laser equipment text semantic feature vector and the laser marking material attribute text semantic feature vector to obtain the laser marking parameter multimodal semantic feature vector.

[0023] It should be understood that feature extraction of the laser marking ambient temperature values ​​collected by the temperature sensor at multiple predetermined time points can extract key features from the raw temperature data that reflect environmental changes during the laser marking process. This can then form a laser marking ambient temperature feature vector, revealing the potential impact and patterns of temperature on the marking process and helping to improve the accuracy and stability of process control. Specifically, from the ambient temperature values ​​at multiple time points, statistical features (such as mean, variance, maximum, minimum, and fluctuation amplitude) can be calculated to summarize the overall temperature distribution. These statistical features can reflect temperature stability and fluctuation. Furthermore, the rate of temperature change (i.e., the slope of the temperature change) is also a crucial feature, as excessively rapid or slow temperature changes can affect the accuracy and effectiveness of laser marking. By extracting these features, a multidimensional feature vector is formed that comprehensively describes the ambient temperature characteristics during the laser marking process.

[0024] Furthermore, laser equipment text parameters typically include power, frequency, pulse width, and scanning speed. These parameters directly impact the accuracy, efficiency, and quality of laser processing. By extracting features from these text parameters, we can uncover potential correlations between equipment status and processing results, thereby optimizing the marking process. Statistical methods (such as mean, standard deviation, maximum, and minimum values) can be used to extract the distribution characteristics of each parameter to reflect the equipment's operational stability and performance fluctuations. For example, the mean and fluctuation range of laser power can indicate whether the equipment maintains stable output, while frequency fluctuations can help determine whether the equipment is operating normally. Secondly, from equipment operation logs or historical parameter data, time series analysis can be used to extract trend features, such as power trends or periodic frequency fluctuations. These trend features help capture the evolution of equipment status and potential fault signals. Ultimately, through these feature extractions, a high- or low-dimensional semantic feature vector for the laser equipment text can be generated, comprehensively describing the equipment's operating status and performance. Furthermore, the properties of the laser marking material (such as material type, surface roughness, thickness, hardness, absorbance, etc.) directly affect the interaction between laser energy and the material. Extracting the characteristic vectors of these material properties helps predict and optimize the marking quality and ensure the stability and accuracy of the marking process.

[0025] Laser marking is a complex, multifactorial process, where both device parameters and material properties jointly determine the marking effect. Relying on information from either aspect alone may not accurately reflect the actual marking state and results. Therefore, correlating these two types of feature vectors can more comprehensively capture the interaction between the device and material during the marking process, thereby forming a comprehensive feature vector that enhances the model's expressiveness and predictive accuracy. Appropriate fusion methods, such as weighted summation, concatenation, or multiplication, can be used to effectively combine these two types of feature vectors. For example, the laser device parameter feature vector and the material property feature vector can be weighted and summed in a certain proportion to produce a fused vector that integrates information about both device performance and material properties. During this fusion process, certain key parameters (such as power, frequency, and surface roughness) may need to be assigned higher weights to highlight their dominant role in the marking effect. Through this association process, the resulting multimodal semantic feature vector of laser marking parameters not only encompasses the laser device operating conditions and the material's physical properties, but also effectively captures the interaction and joint influence of the device and material in actual operation.

[0026] In a specific embodiment of the present application, the laser marking ambient temperature value feature extraction unit 121 includes: arranging the laser marking ambient temperature values ​​at multiple predetermined time points collected by the temperature sensor into a laser marking ambient temperature input vector; passing the laser marking ambient temperature input vector through a laser marking ambient temperature timing encoder to obtain the laser marking ambient temperature feature vector.

[0027] It should be understood that ambient temperature is a critical external parameter in the laser marking process. Its variations can affect the energy transfer of the laser beam and the thermal response of the material. Arranging the ambient temperature values ​​collected by a temperature sensor at multiple predetermined time points into a laser marking ambient temperature input vector constructs an input feature that reflects the temporal variation of ambient temperature. This input vector effectively captures the impact of temperature changes on the laser marking process and helps analyze how temperature fluctuations affect marking quality and stability. Specifically, the temperature values ​​collected by a temperature sensor at multiple predetermined time points typically exhibit certain time series characteristics. Therefore, arranging these temperature values ​​into an input vector preserves the sequential information of the time series, ensuring that the model can identify temperature trends. For example, if a sensor collects temperature data at multiple fixed time points (such as every second or every minute), this temperature data will consist of multiple values ​​arranged sequentially along the time axis. After these values ​​are arranged in chronological order into a vector, each element of the vector represents the temperature value at a specific time point, forming an information vector that includes the time dimension. Next, to improve the model's sensitivity to temperature changes, these temperature values ​​can be preprocessed, such as normalizing or standardizing, to eliminate dimensional differences in temperature data at different time points. For long time series, input vectors can be generated by truncating fixed-length time windows or using sliding windows to ensure consistent input vector lengths, facilitating model training and prediction. By arranging the laser marking ambient temperature values ​​at multiple time points as input vectors, the temporal characteristics of temperature changes can be preserved, allowing the laser marking process to respond more accurately to ambient temperature changes.

[0028] Furthermore, the laser marking ambient temperature input vector is passed through a laser marking ambient temperature time series encoder to obtain a laser marking ambient temperature feature vector. This aims to extract temporal features from the temperature series, helping the model better understand the impact of dynamic changes in ambient temperature on the laser marking process. The laser marking process is a delicate operation that is highly dependent on external conditions (such as ambient temperature). Temperature fluctuations can directly affect the laser beam's energy output, the material's thermal response characteristics, and the final marking effect. Processing the input vector through a time series encoder effectively captures the temporal dependencies and patterns in temperature variations, providing valuable features for subsequent prediction, optimization, and control. Specifically, the laser marking ambient temperature input vector is time series data composed of temperature values ​​collected at multiple time points arranged in chronological order. Because temperature series may contain short-term fluctuations or long-term trends, traditional simple feature extraction methods (such as mean and standard deviation) may not fully reflect the impact of temperature on the marking process. Temporal encoders (such as long short-term memory networks (LSTMs) or gated recurrent units (GRUs)) can effectively process long-term dependencies and temporal patterns in time series data through their internal recurrent structure. The temporal encoder ensures that each temperature value in the input vector is not only related to the temperature at the current point in time, but also takes into account the impact of temperature changes at previous points in time on the current temperature. After processing the temperature input vector through the temporal encoder, the model extracts important temporal features of the temperature sequence (such as fluctuation patterns, periodic changes, and trends) into a low-dimensional feature vector of the laser marking ambient temperature. This feature vector incorporates the dynamic characteristics of the ambient temperature and can serve as input for subsequent tasks (such as marking quality prediction and modeling the impact of ambient temperature on equipment). Through its recurrent structure and memory units, the temporal encoder can capture the potential delay and nonlinear effects of temperature fluctuations on the laser marking process, thereby providing a more accurate representation of the ambient temperature. Specifically, the laser marking ambient temperature input vector is fully connected encoded using the fully connected layer of the laser marking ambient temperature temporal encoder to extract high-dimensional implicit features of the eigenvalues ​​at each position in the laser marking ambient temperature input vector; and the laser marking ambient temperature input vector is one-dimensionally convolutionally encoded using the one-dimensional convolutional layer of the sequence encoder to extract high-dimensional implicit correlation features of the correlation between the eigenvalues ​​at each position in the laser marking ambient temperature input vector.

[0029] In a specific embodiment of the present application, the laser equipment text parameter feature extraction unit 122 includes: segmenting the laser equipment text parameters collected from the database to obtain a laser equipment text parameter word sequence; passing the laser equipment text parameter word sequence through a laser equipment text bidirectional long short-term memory neural network to obtain the laser equipment text semantic feature vector.

[0030] It's understandable that textual parameters for laser equipment typically include information such as model, power, frequency, and spot size. This information may be stored in a database in natural language. Directly analyzing raw text data is not only complex but also difficult to extract useful features. Word segmentation, a crucial step in text preprocessing, breaks down these long texts into their most basic language units, enabling computers to perform subsequent analysis more efficiently. Specifically, textual parameters for laser equipment are typically recorded in databases as sentences or paragraphs, containing multiple terms describing equipment performance and configuration. To convert this textual data into input features that can be used by machine learning models, word segmentation is required to break it down into individual words or phrases. The word segmentation process first involves linguistic analysis of the text to identify its constituent elements, such as words, phrases, and symbols. For Chinese word segmentation, specialized word segmentation tools (such as Jieba and THULAC) may be required. These tools can recognize and process specialized vocabulary, numbers, and units, ensuring accurate word sequences after word segmentation. Through word segmentation, the text parameters of the laser equipment are converted into a sequence of words or phrases with clear meanings. These words can be used as components of the input feature vector for subsequent machine learning models to perform feature extraction and analysis.

[0031] Furthermore, the laser equipment text parameter word sequence is processed through a bidirectional long short-term memory (Bi-LSTM) neural network. This aims to extract richer semantic features from the text sequence, forming a semantic feature vector that comprehensively represents the device parameters. Laser equipment text parameters, such as model, power, and frequency, typically consist of multiple words, and the meaning of these words often depends on their order and context within the text. Therefore, using a bidirectional long short-term memory (Bi-LSTM) network can effectively capture the temporal dependencies between words while simultaneously considering both forward and backward contextual information, thereby extracting semantic features from the text. Specifically, the laser equipment text parameter word sequence is typically a sequence of words arranged in a chronological or logical order, and the order of these words in the text is closely related to the context. Traditional recurrent neural networks (RNNs) can only process sequential data in one direction (usually from left to right). While long short-term memory networks (LSTMs) introduce memory cells to better capture long-term dependencies, they still cannot effectively consider both forward and backward contextual information. To address this issue, the bidirectional LSTM (Bi-LSTM) processes text data simultaneously in both forward and reverse directions, capturing both left-to-right and right-to-left contextual information, thereby more accurately understanding the semantics of each word in the sequence. After Bi-LSTM processing of a laser equipment text parameter word sequence, a vector representing the contextual information of the entire sequence is generated. This vector not only captures the characteristics of the words themselves but also integrates the relationships between each word in the context, thereby comprehensively representing the semantics of the device parameters. Processing the laser equipment text parameter word sequence through a bidirectional long short-term memory neural network (Bi-LSTM) preserves the order of each word in the word sequence while extracting its contextual semantic features. Specifically, the embedding layer of the laser equipment text bidirectional long short-term memory neural network is used to map each laser equipment text parameter word in the laser equipment text parameter word sequence into a word embedding vector to obtain a sequence of laser equipment text parameter word embedding vectors; the converter-based Bert model of the laser equipment text bidirectional long short-term memory neural network is used to perform global context semantic encoding on the sequence of laser equipment text parameter word embedding vectors to obtain multiple laser equipment text parameter feature vectors; and, the multiple laser equipment text parameter feature vectors are cascaded to obtain the laser equipment text semantic feature vector.

[0032] In a specific embodiment of the present application, the laser marking material attribute text parameter feature extraction unit 124 includes: embedding the laser marking material attribute text parameters through a laser marking material attribute text word encoder to obtain a laser marking material attribute text word vector sequence; and passing the laser marking material attribute text word vector sequence through a laser marking material attribute text convolutional neural network based on a multi-scale neighborhood feature extraction module to obtain the laser marking material attribute text semantic feature vector.

[0033] It's understandable that textual parameters for laser marking material properties, such as "hardness," "density," and "reflectivity," often contain specialized terms and numerical values. This information, stored in text form, is difficult to directly utilize effectively in machine learning models. A word embedding encoder can convert this textual data into vectors, thereby providing better input features for the model. Processing the textual parameters for laser marking material properties through the word embedding encoder aims to convert the original textual parameters into a more dense and semantically meaningful numerical representation, namely, a sequence of word vectors for the laser marking material properties text. The core purpose of this process is to map each individual word or phrase into a high-dimensional space through word embedding, effectively capturing and representing the semantic relationships between these words. Specifically, textual parameters for laser marking material properties are typically descriptive text composed of multiple words. Traditional text processing methods (such as TF-IDF based on word frequency) only consider the frequency of word occurrence and fail to capture the semantic relationships between words. A word embedding encoder, however, maps each word into a continuous, high-dimensional space, placing words with similar semantics closer together. In practice, the word embedding encoder for laser marking material property text first converts each word or phrase into a fixed-dimensional vector through an embedding layer. These vectors preserve the semantic information between words. Modern word embedding techniques such as Word2Vec, GloVe, or the more advanced BERT leverage large-scale corpora to learn relationships between words, resulting in a high-dimensional vector that reflects both word meaning and context. Each input text parameter (such as "hardness" or "density") is mapped to a corresponding word vector. The word vectors corresponding to multiple words form a word vector sequence, fully expressing the semantic information of the material property text. Finally, the sequence of word vectors for laser marking material property text generated by the word embedding encoder can be used as input features for machine learning models to analyze the impact of material properties on the laser marking process. These vectors are not only computationally efficient but also better capture the semantic information of material properties, supporting subsequent marking process optimization, quality control, and fault diagnosis.

[0034] Furthermore, after the laser-marked material attribute text is converted into a sequence of word vectors using word embedding technology, while the semantics of each word are captured, to fully understand the semantics of the entire text, features from different perspectives and scales must be extracted. The multi-scale neighborhood feature extraction module is designed to effectively capture the relationships between words and their contextual information at different scales, helping the model better understand the global semantics of the material attribute text. The sequence of word vectors for the laser-marked material attribute text is processed through a convolutional neural network (CNN) based on the multi-scale neighborhood feature extraction module. This leverages the powerful feature extraction capabilities of CNNs, particularly when processing text data with local patterns, to mine deeper semantic information from the text, ultimately generating a feature vector representing the text's semantics. Specifically, the word embedding encoder converts the sequence of word vectors for the laser-marked material attribute text into a series of continuous high-dimensional vectors, each of which contains the semantic information of the word. However, the semantic connections between words and the patterns hidden in the text are often multi-scale, especially when describing material properties, where different levels of information may be present. Traditional convolutional neural networks (CNNs) extract important features through local receptive fields (i.e., capturing local features within a small range). However, when faced with text data containing features at different levels, a single-scale convolution may not be sufficient to capture complex semantic relationships. Therefore, a multi-scale neighborhood feature extraction module is introduced to perform convolution operations on the input word vector sequence at multiple scales, thereby extracting semantic features meaningful to laser-marked material attribute text at different scales. During operation, the convolutional neural network based on the multi-scale neighborhood feature extraction module applies multiple convolution kernels of different sizes to each word vector sequence, extracting short-range, long-range, and even more complex local features. These convolution kernels, using different window sizes, can effectively identify patterns at different levels, such as the features of individual attribute words, relationships between words, and even the contextual information of the entire attribute description. The network's convolutional layers generate multiple feature maps, which are then reduced and aggregated through pooling operations to ultimately produce a high-dimensional semantic feature vector for the laser-marked material attribute text, which comprehensively represents the semantic information of the text. Using a convolutional neural network based on a multi-scale neighborhood feature extraction module to process the word vector sequence of laser marking material properties not only extracts local features but also captures deeper semantic relationships, thereby forming an efficient feature vector rich in semantic information. This feature vector provides more accurate and comprehensive input data for subsequent marking quality prediction, process optimization, and fault diagnosis, and improves the model's understanding of the impact of material properties on laser marking.Specifically, the laser-marked material attribute text convolutional neural network based on the multi-scale neighborhood feature extraction module includes: a first convolutional layer, a second convolutional layer parallel to the first convolutional layer, and a cascade layer connected to the first and second convolutional layers, wherein the first convolutional layer uses a one-dimensional convolution kernel with a first scale, and the second convolutional layer uses a one-dimensional convolution kernel with a second scale. More specifically, the first convolutional layer of the laser-marked material attribute text convolutional neural network based on the multi-scale neighborhood feature extraction module is used to perform one-dimensional convolution encoding on the laser-marked material attribute text word vector sequence to obtain a first-scale feature vector; the second convolutional layer of the multi-scale neighborhood feature extraction module is used to perform one-dimensional convolution encoding on the laser-marked material attribute text word vector sequence to obtain a second-scale feature vector; and the first-scale feature vector and the second-scale feature vector are cascaded to obtain the laser-marked material attribute text semantic feature vector.

[0035] In the aforementioned multi-color laser marking intelligent control system 100 based on multi-dimensional dynamic calibration, the multi-color laser marking device parameter adjustment judgment module 130 is configured to determine whether the multi-color laser marking device parameters require adjustment based on the laser marking ambient temperature feature vector and the multimodal semantic feature vector of the laser marking parameters. It should be understood that by integrating the ambient temperature feature vector and the multimodal semantic feature vector of the laser marking parameters, comprehensive monitoring and adjustment of the environmental and process parameters during the multi-color laser marking process can be achieved, ensuring that the device consistently maintains optimal performance under various operating conditions, resulting in more accurate and stable marking results.

[0036] Figure 3 This is a block diagram of a multi-color laser marking device parameter adjustment judgment module in a multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to an embodiment of the present application. Figure 3 As shown, in a specific embodiment of the present application, the multi-color laser marking device parameter adjustment judgment module 130 includes: a laser marking device feature fusion unit 131, which is used to fuse the laser marking ambient temperature feature vector and the laser marking parameter multimodal semantic feature vector to obtain a laser marking device adjustment judgment feature vector; a laser marking device feature optimization unit 132, which is used to perform topological space constraints on the laser marking device adjustment judgment feature vector based on cross-domain transfer learning to obtain an optimized laser marking device adjustment judgment feature vector; a laser marking device adjustment parameter adjustment classification judgment unit 133, which is used to pass the optimized laser marking device adjustment judgment feature vector through a classifier to obtain a classification result, and the classification result is used to determine whether the multi-color laser marking device parameters need to be adjusted.

[0037] It should be understood that the ambient temperature feature vector contains numerical information about the current ambient temperature, which reflects the potential impact of temperature on the marking effect. The multimodal semantic feature vector of the laser marking parameters, on the other hand, integrates multidimensional marking process information (such as power, speed, and focal length), each dimension of which reflects the process's distinct contribution to marking quality. By fusing these two feature vectors, the relationship between temperature and process parameters, as well as their combined impact on marking quality, can be more accurately captured. By integrating the optimized laser marking equipment adjustment judgment feature vector, the system can monitor current environmental conditions and process settings in real time and automatically determine whether equipment parameter adjustments are necessary, thereby optimizing marking results, improving production efficiency, and avoiding quality fluctuations.

[0038] In particular, considering that ambient temperature data is time-series numerical data, while equipment parameters and material properties are textual information, the temperature data is converted into feature vectors using a time-series encoder. These features primarily reflect the temporal trend of temperature, but lack deep semantic connections with equipment operation or material properties. Equipment parameters and material properties, on the other hand, are extracted using natural language processing methods, using bidirectional LSTM and convolutional neural networks, respectively, to extract semantic features. These focus on the semantics, syntax, and context of text, which is completely different from the time-series nature of temperature data. When fusing these data from different sources, due to significant differences in dimensionality, information type, and semantic level, the fused feature vectors for laser marking equipment adjustment judgment may not fully preserve the unique information of each data type. Furthermore, while the textual features of equipment and materials can express complex semantic relationships, because these two types of text data are processed separately and their embedding vectors cannot directly reflect the real-time relationship with the external physical environment, such as temperature, the final feature fusion may not effectively strengthen the intrinsic connections between data points. Therefore, although the fused laser marking equipment adjustment judgment feature vector contains multimodal information, due to the heterogeneity of the information and the multi-level semantic differences, it may not be able to accurately capture and express the complex interactions between factors such as temperature, equipment, and material properties while maintaining the original characteristics of various types of data. This difference makes it difficult for the fused feature vector to comprehensively improve the intrinsic connection between data points, thereby affecting the accuracy of equipment adjustment judgment. Therefore, in the technical solution of the present application, the laser marking equipment adjustment judgment feature vector is subjected to topological space constraints based on cross-domain transfer learning to obtain an optimized laser marking equipment adjustment judgment feature vector.

[0039] Among them, the laser marking equipment adjustment judgment feature vector is subjected to topological space constraints based on cross-domain transfer learning to obtain an optimized laser marking equipment adjustment judgment feature vector, including: extracting the target parameter matrix of the classifier; performing node decomposition on the target parameter matrix in units of row vectors to obtain a set of target parameter node coding vectors; using each target parameter node coding vector in the set of target parameter node coding vectors as a wandering topological space, and performing topological space constraints on the laser marking equipment adjustment judgment feature vector to obtain a set of constrained laser marking equipment adjustment judgment feature vectors; and calculating the positional mean vector of the set of constrained laser marking equipment adjustment judgment feature vectors to obtain the optimized laser marking equipment adjustment judgment feature vector.

[0040] Among them, each target parameter node coding vector in the set of target parameter node coding vectors is used as the wandering topological space, and the laser marking equipment adjustment judgment feature vector is topologically constrained to obtain a set of constrained laser marking equipment adjustment judgment feature vectors, including: multiplying the laser marking equipment adjustment judgment feature vector and the transposed vector of the target parameter node coding vector, and calculating the natural exponential function value of the multiplication result to obtain a weighted exponential response weight; calculating the Euclidean distance between the laser marking equipment adjustment judgment feature vector and the target parameter node coding vector to obtain a node coding distance value; performing point multiplication on the node coding distance value and the laser marking equipment adjustment judgment feature vector, and calculating the natural exponential function value for each eigenvalue of the vector after the point multiplication to obtain a laser marking equipment distance guidance index feature vector; multiplying the weighted exponential response weight and the laser marking equipment distance guidance index feature vector to obtain a constrained laser marking equipment adjustment judgment feature vector.

[0041] The topological space constraint based on cross-domain transfer learning is performed on the laser marking equipment adjustment judgment feature vector to obtain an optimized laser marking equipment adjustment judgment feature vector, which is expressed by the following optimization formula: in, represents the target parameter matrix, The first, second, and third sets of target parameter node encoding vectors are represented , target parameter node encoding vector, represents the transpose of a vector, represents the adjustment judgment feature vector of the laser marking equipment, represents matrix multiplication, Indicates point multiplication by position, Represents the calculation vector and vector The Euclidean distance between The first one represents the set of feature vectors of the laser marking equipment adjustment judgment after constraint After the constraints, the laser marking equipment adjusts the judgment feature vector, Represents the total number of sets of feature vectors for adjusting the laser marking equipment after constraints, Represents the optimized laser marking equipment adjustment judgment feature vector.

[0042] In the technical solution of this application, a topological spatial constraint based on cross-domain transfer learning is applied to the feature vectors used to adjust the laser marking equipment. This process first extracts the key parameters used for decision-making from the trained classifier. These parameters form a matrix in a high-dimensional space, with each row representing the weight or influencing factor on a different dimension. This target parameter matrix provides insight into the location and shape of the model's decision boundary, and further infers which input features are most critical to the prediction results.

[0043] Next, the target parameter matrix is ​​node-decomposed into row vectors to obtain a set of target parameter node encoding vectors. Here, each row vector is treated as a node in the graph theory, which means that each set of parameters is now regarded as an entity with potential connectivity. This transformation allows the application of methods from graph theory and network science to explore the interactions between features. The node encoding vector not only carries information about the original parameters, but also implies knowledge about the topological structure of the entire system. Node decomposition further reveals the intrinsic connection pattern or structure of the data, allowing the optimized feature vector to better adapt to new task requirements.

[0044] Then, each target parameter node encoding vector in the set of target parameter node encoding vectors is used as a topological space for walking, and topological space constraints are respectively applied to the laser marking device adjustment judgment feature vectors to obtain a set of constrained laser marking device adjustment judgment feature vectors. "Walking" in the topological space defined by the node encoding vector is actually simulating an exploratory process, the purpose of which is to find feature transformations that can best maintain the structural characteristics of the original data. Each step determines the position of the next step based on the probability distribution of the current state. The topological space constraints ensure that even in different contexts, the feature representation retains certain invariances. At the same time, it can also promote cross-domain transfer learning because it emphasizes the universal relationship between features rather than the details of a specific field. In this way, the reconstruction of the feature space is achieved, making the optimized feature vector more compact and having better generalization capabilities.

[0045] Finally, the positional mean vector of the set of constraint-based laser marking device adjustment judgment feature vectors is calculated to obtain the optimized laser marking device adjustment judgment feature vector. Calculating the mean vector is a statistical aggregation method used to synthesize the optimal solution from multiple perspectives. The idea behind this step is to reduce the deviation caused by a single estimate by fusing the information provided by different sample points. The averaging process is equivalent to performing a soft voting, which enhances the expressiveness of common features and makes the optimized feature vector more stable and reliable.

[0046] Furthermore, the optimized laser marking equipment adjustment judgment feature vector is a composite vector that combines the ambient temperature feature vector and the multimodal semantic feature vector of the laser marking parameters. It contains comprehensive information about environmental conditions and process parameters. These features reflect the current equipment operating status and the risk of poor marking results due to environmental changes or improper parameter settings. However, because this information is often high-dimensional and complex, relying solely on manual analysis makes it difficult to achieve efficient real-time monitoring and adjustments. Therefore, a machine learning model (classifier) ​​is needed to analyze these features and make appropriate judgments. Classifiers, particularly commonly used algorithms such as decision trees, support vector machines (SVMs), and neural networks, can learn from historical data how environmental factors and equipment parameters affect marking quality and establish corresponding rules or models. When a new feature vector is input, the classifier uses the trained model to determine whether equipment parameter adjustment is necessary. The classification result is usually "adjustment required" or "no adjustment required." Based on the predicted output, the system can automatically or prompt the operator to make appropriate adjustments to the laser equipment parameters. For example, if the classifier determines that the ambient temperature is too high and the current power setting is insufficient, the system may recommend increasing the power or reducing the scanning speed to ensure stable color rendering. In this way, the laser marking equipment can intelligently determine whether parameter adjustments are necessary based on real-time environmental and process information, avoiding the limitations of human intervention and improving production efficiency and consistency of marking quality. This classifier-driven intelligent adjustment mechanism not only automatically adapts to changing operating environments but also enables precise control during the multi-color laser marking process, maximizing the quality and accuracy of the final mark.

[0047] In summary, the embodiment of the present application first obtains the laser marking ambient temperature values ​​at multiple predetermined time points collected by the temperature sensor, the laser equipment text parameters and the laser marking material attribute text parameters collected by the database, and then uses deep learning technology to perform feature extraction and correlation analysis on the three. Finally, the classification results are obtained through the classifier to determine whether the parameters of the multi-color laser marking equipment need to be adjusted, thereby improving the intelligence level of the laser marking process, and then improving production efficiency, reducing manual intervention and debugging difficulty, reducing production costs, and ensuring the reliability and efficiency of the production process.

[0048] In another specific embodiment of the present application, while traditional marking machines can perform color marking, their energy efficiency, accuracy, and adaptability are poor, making them unable to meet modern demands for high precision and diversity. To achieve efficient, multi-color laser display across the full spectrum, another specific embodiment of the present application proposes a novel multi-dimensional dynamic calibration system. This system, combining intelligent feedback adjustment with high-precision scanning technology, overcomes the bottlenecks of traditional processes. Specifically, another specific embodiment of the present application utilizes a full-spectrum laser source, an intelligent two-dimensional high-speed galvanometer, a dynamic grating system, and an adaptive control algorithm to achieve efficient, multi-color marking on the surfaces of various materials. The system also features remote control, automatic fault detection, and multi-level optical anti-counterfeiting capabilities. The laser source module utilizes a tunable full-spectrum laser, supporting broadband operation from ultraviolet to infrared. The laser source module has a wavelength range of 200 nm to 1064 nm, and a power adjustment range of 1 mW to 200 W. The fiber optic transmission module utilizes efficient fiber coupling to reduce energy loss and ensure stable transmission. The intelligent high-speed galvanometer system includes X- and Y-axis galvanometers with a response speed of up to 1 ms and an accuracy of 0.01 degrees. The dynamic grating system uses an electrically controlled grating, enabling real-time frequency and direction adjustment. The F-Theta objective lens assembly enables precise scanning of the laser beam in the focal plane. The adaptive feedback system uses real-time sensors to monitor temperature, humidity, and vibration, dynamically adjusting the beam path and power. The intelligent control module supports remote monitoring and automatic fault detection, and operating parameters can be adjusted in real time via PC software.

[0049] Figure 4 Schematic diagram of the overall architecture of a multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to an embodiment of the present application. Figure 5 Schematic diagram of grating dynamic adjustment of the multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to an embodiment of the present application. Figure 4 and Figure 5 As shown, in another specific embodiment of the present application, a PC is included, and the PC signal is connected to a laser control console and a motor control console. The PC sends a control signal to the laser through the laser control console to control the laser to emit a corresponding laser beam. The PC is connected to the motor signal through the motor control console, and the control signal sent drives the motor to operate, driving the grating and the two-dimensional high-speed galvanometer to adjust to the corresponding angle. The pattern to be marked is recorded on the PC end, and the scanning path and grating frequency of the galvanometer are optimized by simulating the marking process.

[0050] Specifically, a radiator is provided on the outside of the laser, and the radiator is a fan radiator. The fan radiator is used to solve the problem of severe temperature efficiency and high heat generation of the laser, so as to improve the overall processing efficiency of the marking machine. Located at the end of the laser light speed, a marking workpiece is provided. Located between the laser and the marking workpiece, an optical fiber, a grating, a convex lens, a two-dimensional high-speed galvanometer and an F-Theta objective lens are sequentially provided from the laser to the marking workpiece. The grating and the two-dimensional high-speed galvanometer are connected to a motor, and the angles of the grating and the two-dimensional high-speed galvanometer can be adjusted by the motor.

[0051] Specifically, the optical fiber can couple the laser beam. The laser beam is emitted from the laser and further enhances its optical properties through the optical fiber coupling. The fiber-coupled laser beam is then irradiated onto the grating.

[0052] Specifically, the grating is installed on a workbench and controlled by a motor. When it rotates to different positions, gratings of different frequencies are aligned with the laser beam, so that the grating stripes irradiated on the workpiece also have different spatial frequencies; at the same time, the rotation of the optical head makes the captured grating have different grating orientations.

[0053] The grating directs the laser beam onto a convex lens, which then focuses the laser beam onto a two-dimensional high-speed galvanometer. The convex lens's function is to converge the light emitted from the laser. Without this lens, the laser would emit divergent light, reducing its energy.

[0054] Specifically, the two-dimensional high-speed galvanometer includes an X-axis galvanometer and a Y-axis galvanometer, which are connected to the output end of the motor, and the angle and direction of the X-axis galvanometer and the Y-axis galvanometer are adjusted by the forward or reverse rotation of the motor.

[0055] The 2D high-speed galvanometer is the actuator of the laser marking machine's motion. It employs specific techniques to deflect the laser beam's propagation direction according to specific rules. Generally, this is achieved through physical effects such as reflection and diffraction. The laser beam passes through the 2D high-speed galvanometer, completing the optical scanning process. The 2D high-speed galvanometer is a core component of laser marking. During the laser marking process, the 2D high-speed galvanometer's motion is controlled to allow arbitrary two-dimensional movement of the workpiece surface.

[0056] Next, a two-dimensional high-speed galvanometer will direct the laser beam onto the F-Theta objective lens, which interferes with and exposes the workpiece. The rotation of the high-speed two-dimensional galvanometer, combined with the frequency selection of different gratings, can engrave color-changing graphics or text on the surface of the workpiece. The laser frequency λ is 532nm, k=±1, and d is the grating constant. The so-called grating constant means that when the number of light rays within 1mm is n, the distance between each two rays is d=1 / n, and the unit is micrometer (μm). Commonly used gratings are 300dpi, 400dpi, and 500dpi, and their grating constants are 3.3μm, 2.5μm, and 2μm, respectively.

[0057] The grating is controlled by a motor, and its rotation to different positions produces different frequencies of the so-called grating alignment laser beam, resulting in different spatial frequencies of the grating stripes irradiated on the workpiece. Simultaneously, the rotation of the optical head causes the captured grating to have different grating orientations. The rotation of the two-dimensional galvanometer exposes each point on the workpiece, forming a light-variable image composed of pixel gratings with different grating orientations and spatial frequencies. This allows the engraving of color-changing graphics or text, creating multi-color graphics.

[0058] Specifically, the multi-color graphics are formed by a tiny pixel grating matrix. Under different observation angles, pixel grating groups with the same characteristics will form specific graphics, and pixel gratings with different characteristics will present graphics with different effects. The optical parameters describing the optically variable image are: position: set values ​​of the X-axis and Y-axis, diffraction direction of the pixel grating: set value of the grating angle, density distribution of the pixel grating: set value of the color, and luminous efficiency of the pixel: set value of the brightness.

[0059] The color and shape of the optically variable image will change with the movement of light or changes in the viewing direction. The diffraction of light on each pixel grating satisfies the grating equation, namely:

[0060] That is to say, the color of the marking pattern will change with the movement of light and the change of the observer's orientation, which has a very good color rendering effect. This will better solve the problem of poor color combination of previous laser marking machines.

[0061] Among them, the color-changing laser marking system uses the interference of laser beams on the surface of the material. The rotation of the grating in the system changes the orientation of the interference stripes of the interference light spots, and reasonably arranges the orientation of the interference stripes of each light spot in the graphic, so that the recorded graphics can reproduce beautiful text or graphics under white light.

[0062] Figure 6 This is a flow chart of the intelligent control module of the multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to an embodiment of the present application. Figure 6 As shown, in another specific embodiment of the present application, first, a printed graphic is input on a PC and marking is simulated on the PC. At the end of the simulation, the parameters of the marking machine are saved, mainly including the parameter settings of the laser control console and the parameter settings of the motor. The parameters of the laser control console mainly involve beam quality factor, far-field divergence angle, polarization direction, polarization ratio, pulse energy, pulse width, beam directivity and power, etc. The determination of the above parameters needs to be obtained by simulating the characteristics of the metal device on the PC. The parameters of the motor control console include the motor parameters for controlling the rotation direction of the grating and the motor parameters for controlling the scanning direction of the two-dimensional high-speed galvanometer.

[0063] By controlling the rotation of the pixel grating through a motor, grating orientations of different angles and directions can be obtained. When it is rotated to different positions, pixel gratings of different frequencies will be aligned with the laser beam, thereby making the grating stripes irradiated on the metal workpiece diverse.

[0064] The 2D high-speed galvanometer is controlled by a motor. Its response time is 1ms when controlling the X- and Y-axis galvanometers. Control is via analog signals from a PC. The coordinate values ​​of each point in the graph are converted into corresponding voltage values. The rotation angle of the X- and Y-axis galvanometers is linearly related to the voltage signal, with a voltage change of 0.5V per degree of rotation. The galvanometer control voltage signal is ±10V, resulting in a rotation angle of ±20°. Polarization rotation is combined with the 2D high-speed galvanometer to transfer the laser beam to an F-Theta objective. With a normal focusing lens, parallel light at infinity forms an image on the focal plane. However, the ideal image height and scanning angle are often not linear, making constant-speed scanning impossible. This can affect laser marking accuracy. The F-Theta objective effectively addresses this issue, enabling linear scanning of the incident beam deflected at a constant angular velocity on the focal plane.

[0065] In this way, precise control of multi-color dynamic marking can be achieved, adaptive multi-frequency adjustment can be performed, and real-time sensors and control algorithms can ensure the long-term stability of the system. Laser interference and multi-dimensional scanning technology can be integrated to perform multi-level anti-counterfeiting functions, making it easy to maintain and upgrade and adapt to different industrial scenarios.

[0066] As described above, the multi-color laser marking intelligent control system 100 based on multi-dimensional dynamic calibration according to the embodiment of the present application can be implemented in various terminal devices. In one example, the multi-color laser marking intelligent control system 100 based on multi-dimensional dynamic calibration can be integrated into the terminal device as a software module and / or hardware module. For example, the multi-color laser marking intelligent control system 100 based on multi-dimensional dynamic calibration can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the multi-color laser marking intelligent control system 100 based on multi-dimensional dynamic calibration can also be one of the many hardware modules of the terminal device.

[0067] Alternatively, in another example, the multi-color laser marking intelligent control system 100 based on multi-dimensional dynamic calibration and the terminal device can also be separate devices, and the multi-color laser marking intelligent control system 100 based on multi-dimensional dynamic calibration can be connected to the terminal device through a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

Claims

1. A multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration, characterized in that: include: A multi-color laser marking data acquisition module, used to acquire the laser marking environment temperature values ​​at multiple predetermined time points acquired by the temperature sensor, the laser equipment text parameters acquired by the database, and the laser marking material property text parameters; A multi-color laser marking data extraction module, used to extract a laser marking ambient temperature feature vector and a laser marking parameter multimodal semantic feature vector from the laser marking ambient temperature values ​​at multiple predetermined time points collected by the temperature sensor, the laser equipment text parameters collected by the database, and the laser marking material attribute text parameters; The multi-color laser marking equipment parameter adjustment judgment module is used to judge whether the multi-color laser marking equipment parameters need to be adjusted based on the laser marking environment temperature feature vector and the laser marking parameter multimodal semantic feature vector.

2. The multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to claim 1 is characterized in that: The multi-color laser marking data extraction module comprises: A laser marking environment temperature value feature extraction unit, used for performing feature extraction on the laser marking environment temperature values ​​at a plurality of predetermined time points collected by the temperature sensor to obtain the laser marking environment temperature feature vector; A laser equipment text parameter feature extraction unit, used for extracting features from the laser equipment text parameters collected from the database to obtain a laser equipment text semantic feature vector; A laser marking material attribute text parameter feature extraction unit, used for extracting features from the laser marking material attribute text parameters to obtain a laser marking material attribute text semantic feature vector; The laser marking parameter multimodal feature association unit is used to associate the laser equipment text semantic feature vector with the laser marking material attribute text semantic feature vector to obtain the laser marking parameter multimodal semantic feature vector.

3. The multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to claim 2 is characterized in that: The laser marking environment temperature value feature extraction unit comprises: Arranging the laser marking environment temperature values ​​at a plurality of predetermined time points collected by the temperature sensor as a laser marking environment temperature input vector; The laser marking environment temperature input vector is passed through a laser marking environment temperature timing encoder to obtain the laser marking environment temperature feature vector.

4. The multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to claim 3 is characterized in that: The laser device text parameter feature extraction unit comprises: Segmenting the laser equipment text parameters collected from the database to obtain a laser equipment text parameter word sequence; The laser equipment text parameter word sequence is passed through a laser equipment text bidirectional long short-term memory neural network to obtain the laser equipment text semantic feature vector.

5. The multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to claim 4 is characterized in that: The laser marking material attribute text parameter feature extraction unit comprises: Embedding the laser marking material attribute text parameters through a laser marking material attribute text word encoder to obtain a laser marking material attribute text word vector sequence; The laser-marked material attribute text word vector sequence is passed through a laser-marked material attribute text convolutional neural network based on a multi-scale neighborhood feature extraction module to obtain the laser-marked material attribute text semantic feature vector.

6. The multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to claim 5 is characterized in that: The multi-color laser marking equipment parameter adjustment judgment module includes: A laser marking equipment feature fusion unit, used to fuse the laser marking environment temperature feature vector and the laser marking parameter multimodal semantic feature vector to obtain a laser marking equipment adjustment judgment feature vector; A laser marking equipment feature optimization unit, used for performing topological space constraints on the laser marking equipment adjustment judgment feature vector based on cross-domain transfer learning to obtain an optimized laser marking equipment adjustment judgment feature vector; The laser marking equipment adjustment parameter adjustment classification judgment unit is used to pass the optimized laser marking equipment adjustment judgment feature vector through a classifier to obtain a classification result, and the classification result is used to judge whether the multi-color laser marking equipment parameters need to be adjusted.

7. The multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to claim 6 is characterized in that: The laser marking equipment feature optimization unit comprises: Extract the target parameter matrix of the classifier; Decomposing the target parameter matrix into nodes in units of row vectors to obtain a set of target parameter node encoding vectors; Taking each target parameter node encoding vector in the set of target parameter node encoding vectors as the walking topological space, topological space constraints are respectively applied to the laser marking device adjustment judgment feature vectors to obtain a set of constrained laser marking device adjustment judgment feature vectors; and The position mean vector of the set of the constrained laser marking equipment adjustment judgment feature vectors is calculated to obtain the optimized laser marking equipment adjustment judgment feature vector.

8. The multi-color laser marking intelligent control system based on multi-dimensional dynamic calibration according to claim 7 is characterized in that: Taking each target parameter node encoding vector in the set of target parameter node encoding vectors as the walking topological space, topological space constraints are respectively performed on the laser marking device adjustment judgment feature vectors to obtain a set of constrained laser marking device adjustment judgment feature vectors, including: After multiplying the laser marking device adjustment judgment feature vector and the transposed vector of the target parameter node encoding vector, a natural exponential function value of the multiplication result is calculated to obtain a weighted exponential response weight; Calculating the Euclidean distance between the laser marking device adjustment judgment feature vector and the target parameter node coding vector to obtain a node coding distance value; Performing a dot multiplication on the node coding distance value and the laser marking device adjustment judgment feature vector, and calculating a natural exponential function value for each feature value of the vector after the dot multiplication to obtain a laser marking device distance guidance index feature vector; The weighted index response weight and the laser marking device distance guidance index feature vector are multiplied to obtain a constrained laser marking device adjustment judgment feature vector.

Images