Print control method and system

Abstract

This invention relates to the field of printing control technology and discloses a printing control method and system. The method includes: preprocessing the text image of a sign to obtain a sequence of characters to be converted, and generating an actuator drive instruction group based on the sequence of characters to be converted; synchronously driving the printing actuator to perform Braille dot imprinting operations while simultaneously acquiring imprinting force and position deviation signals in real time to obtain actuator operating status feedback data; dynamically adjusting parameters through a dual-channel attention control network to obtain a first actuator control signal and the completed Braille dot matrix; performing optical detection and geometric dimension measurement to obtain Braille dot forming quality deviation data; and performing gradient analysis and compensation on the first actuator control signal to obtain a second actuator control signal for the next printing process. This invention has good self-learning capabilities, ensuring high precision, high efficiency, and high stability in the production of Braille signs.

Description

Technical Field

[0001] This invention relates to the field of printing control technology, and in particular to a printing control method and system. Background Technology

[0002] Traditional Braille sign production relies primarily on manual embossing or simple mechanical embossing methods. These methods are not only inefficient but also struggle to guarantee the accuracy and consistency of the Braille dots, exhibiting significant limitations, particularly in mass production and handling complex text. Furthermore, existing automated Braille production equipment generally suffers from technical challenges in image and text recognition, such as low accuracy in recognizing complex backgrounds and high error rates in converting polyphonic characters. These issues severely impact the quality and production efficiency of Braille signs. Summary of the Invention

[0003] This invention provides a printing control method and system. This invention has good self-learning ability, ensuring high precision, high efficiency and high stability in the production of Braille signs.

[0004] The first aspect of the present invention provides a printing control method, the printing control method comprising: The text image of the sign is preprocessed to obtain a sequence of characters to be converted, and an actuator drive instruction set is generated based on the sequence of characters to be converted; Based on the actuator drive instruction group, the printing actuator is synchronously driven to perform Braille dot imprinting operation, while the imprinting force and position deviation signals are collected in real time to obtain actuator operating status feedback data. The actuator operating status feedback data is input into a dual-channel attention control network for dynamic parameter adjustment to obtain the first actuator control signal and the completed Braille dot matrix; Optical inspection and geometric dimension measurement are performed on the completed Braille dot matrix to obtain Braille dot forming quality deviation data; Based on the Braille dot forming quality deviation data, gradient analysis and compensation are performed on the first actuator control signal to obtain the second actuator control signal for the next printing process.

[0005] In conjunction with the first aspect, in a first implementation of the first aspect of the present invention, the step of preprocessing the signboard text image to obtain a sequence of characters to be converted, and generating an actuator drive instruction set based on the sequence of characters to be converted, includes: Sobel gradient calculation and edge intensity extraction are performed on the text image of the sign to obtain image edge feature data; The edge feature data of the image and the text image of the sign are weighted and fused and Gaussian filtered to obtain an edge-enhanced fused image; The edge-enhanced fused image is subjected to adaptive threshold segmentation to obtain a binarized text region image; The binarized text region image is input into a multi-scale recognition network for residual connection feature extraction and sequence decoding to obtain the text character sequence to be converted. Semantic context analysis and Braille encoding mapping are performed on the character sequence to be converted to obtain the actuator drive instruction set.

[0006] In conjunction with the first aspect, in a second implementation of the first aspect of the present invention, the step of inputting the binarized text region image into a multi-scale recognition network for residual connection feature extraction and sequence decoding to obtain the text character sequence to be converted includes: The binarized text region image is simultaneously distributed to three parallel feature extraction branches in the multi-scale recognition network. The small-scale branch of the three parallel feature extraction branches is used to capture text detail edge features, the medium-scale branch of the three parallel feature extraction branches is used to extract text local structural features, and the large-scale branch of the three parallel feature extraction branches is used to obtain text global features, resulting in three sets of multi-scale original feature maps with different receptive fields. Deep feature learning and channel concatenation are performed on the three sets of original feature maps with different receptive fields at multiple scales to obtain a multi-scale fused feature map. The feature vectors of the multi-scale fused feature map are then serialized and rearranged to obtain a feature sequence. The feature sequence is input into the decoder for forward computation, and the merging rules remove duplicate characters and whitespace markers to obtain the text character sequence to be converted.

[0007] In conjunction with the first aspect, in a third implementation of the first aspect of the present invention, the step of performing semantic context analysis and Braille encoding mapping on the character sequence to be converted to obtain an executor drive instruction set includes: After extracting context information for each target character in the text character sequence to be converted, the information is input into the BERT pre-trained model for context semantic vector encoding to obtain the character context representation vector. The character context representation vector is input into the polyphonic character disambiguation module, and a unique phonetic annotation is determined from multiple candidate pronunciations of the polyphonic character disambiguation module to obtain the disambiguated character pronunciation sequence. Based on the disambiguated character pronunciation sequence, a pre-constructed Chinese character-Braille mapping matrix is ​​searched to obtain the corresponding six-dot Braille code. Then, each six-dot Braille code is converted into Braille dot matrix spatial distribution data in a two-dimensional spatial coordinate system through the Braille dot coordinate transformation function. Based on the spatial distribution data of the Braille dot matrix, an actuator drive instruction set containing dot coordinates, imprint height, imprint diameter, and imprint force is generated.

[0008] In conjunction with the first aspect, in the fourth implementation of the first aspect of the present invention, the step of synchronously driving the printing actuator based on the actuator drive instruction group to perform Braille dot imprinting operation while simultaneously acquiring imprinting force and position deviation signals in real time to obtain actuator operating status feedback data includes: According to the point coordinates, imprint height, imprint diameter, and imprint force in the actuator drive instruction group, the printing actuator's imprint head is driven to move to the target position and perform Braille dot imprinting operation; During the ongoing Braille dot formation process, the real-time imprinting force between the imprint head and the substrate is continuously sampled and monitored using a piezoresistive sensor to obtain imprinting force time-series data. Simultaneously, a photoelectric encoder is used to measure and calculate the actual position of the printing actuator in the x-axis and y-axis directions in real time to obtain position error data; By adding the spatial coordinates and time markers of the current imprint point to the imprint force timing data and the position error data, the actuator operating status feedback data is obtained.

[0009] In conjunction with the first aspect, in the fifth implementation of the first aspect of the present invention, the step of inputting the actuator operating status feedback data into a dual-channel attention control network for dynamic parameter adjustment to obtain a first actuator control signal and the completed Braille dot matrix includes: The actuator operating status feedback data is separated into pressure channel input data and position channel input data according to data type. The pressure channel input data includes imprint force timing data, and the position channel input data includes position error data in the x and y directions. The pressure channel input data and the position channel input data are respectively input into a dual-channel separate control network to calculate the attention weights and obtain the dual-channel attention weight coefficients. The attention control factor is obtained by weighting and fusing the dual-channel attention weight coefficients. The compensation control quantity is calculated based on the fused attention control factor, and a first actuator control signal is generated based on the actuator dynamics model and the compensation control quantity. Based on the control signal of the first actuator, the printing actuator is driven to complete the imprinting and forming operation of the current Braille dot and update the overall printing status to obtain the completed Braille dot matrix.

[0010] In conjunction with the first aspect, in a sixth implementation of the first aspect of the present invention, the step of calculating a compensation control quantity based on the fused attention control factor, and simultaneously generating a first actuator control signal based on the actuator dynamics model and the compensation control quantity, includes: By performing a difference calculation between the target position coordinates of the current imprint point and the actual measured position coordinates, and simultaneously performing a difference calculation between the target imprint force and the actual imprint force, a current tracking error vector containing position tracking error and pressure tracking error is obtained. Based on the current tracking error vector, the proportional term, integral term and derivative term are calculated using the PID control law to obtain the basic PID control quantity; The fused attention control factor is multiplied by a preset compensation gain coefficient to generate a compensation control quantity; The actuator dynamics model established based on the actuator mass matrix, damping matrix, and stiffness matrix is ​​used to perform dynamic constraint verification on the PID basic control quantity and the compensation control quantity to obtain the adjustment control quantity under dynamic constraints. Based on the adjustment control quantity under the aforementioned dynamic constraints and the current actuator state information, the control signal is amplitude modulated and timing adjusted to obtain the first actuator control signal.

[0011] In conjunction with the first aspect, in the seventh implementation of the first aspect of the present invention, the step of performing optical inspection and geometric dimension measurement on the completed Braille dot matrix to obtain Braille dot forming quality deviation data includes: A laser displacement sensor is used to optically detect the convex height of each Braille dot in the completed Braille dot matrix, and a Braille dot height measurement dataset is obtained. Image acquisition and contour boundary extraction are performed on the completed Braille dot matrix, and the Braille dot diameter measurement dataset is calculated; The Braille dot height measurement dataset and the Braille dot diameter measurement dataset are respectively compared with the standard height value and the standard diameter value to obtain a geometric deviation parameter set, and Braille dot forming quality deviation data are generated based on the geometric deviation parameter set.

[0012] In conjunction with the first aspect, in the eighth implementation of the first aspect of the present invention, the step of performing gradient analysis and compensation on the first actuator control signal based on the Braille dot forming quality deviation data to obtain the second actuator control signal for the next printing process includes: The time derivative of the Braille dot forming quality deviation data is calculated to obtain deformation gradient data, and a sliding window analysis is performed based on the deformation gradient data to obtain the result of the actuator performance degradation trend judgment. When the result of the actuator performance degradation trend judgment exceeds the preset degradation threshold range, the fault diagnosis process is initiated and an actuator performance compensation strategy selection instruction is generated. According to the actuator performance compensation strategy, the control gain parameter in the first actuator control signal is adaptively adjusted by the selection instruction to obtain the target control parameter set. The target control parameter set is combined with the actuator remaining life prediction data to generate the second actuator control signal for the next printing process.

[0013] A second aspect of the present invention provides a printing control system, the printing control system comprising: The preprocessing module is used to preprocess the text image of the sign to obtain the character sequence of the text to be converted, and to generate an actuator drive instruction group based on the character sequence of the text to be converted. The status feedback module is used to synchronously drive the printing actuator to perform Braille dot imprinting operation based on the actuator drive instruction group, and collect imprinting force and position deviation signals in real time to obtain actuator operation status feedback data. The parameter adjustment module is used to input the actuator operating status feedback data into the dual-channel attention control network for dynamic parameter adjustment, so as to obtain the first actuator control signal and the completed Braille dot matrix; The measurement module is used to perform optical detection and geometric dimension measurement on the completed Braille dot matrix to obtain Braille dot forming quality deviation data; The compensation module is used to perform gradient analysis and compensation on the first actuator control signal based on the Braille dot forming quality deviation data, so as to obtain the second actuator control signal for the next printing process.

[0014] Compared with existing technologies, this invention has the following advantages: By employing Sobel operator edge enhancement preprocessing technology combined with a multi-scale recognition network, the accuracy of character recognition in complex backgrounds is significantly improved; the polyphonic character disambiguation module based on BERT semantic analysis effectively reduces the Braille conversion error rate; the dual-channel attention control network achieves precise coordinated control of pressure and position signals, significantly improving printing accuracy compared to existing open-loop control methods; the joint detection of laser displacement sensor and vision sensor establishes a complete real-time quality monitoring system; the fault diagnosis algorithm based on deformation gradient realizes predictive maintenance, shifting from passive maintenance to proactive prediction and extending equipment life; this invention constructs a fully automated technology chain from image recognition, semantic understanding, precision control to quality inspection, and achieves adaptive control optimization through gradient analysis and compensation mechanisms, possessing good self-learning capabilities, ensuring high precision, high efficiency, and high stability in Braille sign production. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] The structures, proportions, sizes, etc., shown in the accompanying drawings of this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed in the specification, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0017] Figure 1 This is a flowchart illustrating the printing control method provided in an embodiment of the present invention; Figure 2 This is a schematic block diagram of the printing control system provided in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0020] It should also be understood that the terminology used in this specification is for the purpose of describing preferred embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0021] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items, and all possible combinations, and includes such combinations. See also Figure 1 One embodiment of the printing control method in this invention includes: Step 100: Preprocess the sign text image to obtain the character sequence to be converted, and generate an actuator drive instruction group based on the character sequence to be converted; It is understood that the executing entity of this invention can be a printing control system, a terminal, or a server; no specific limitation is made here. This embodiment of the invention will be described using a server as an example.

[0022] Specifically, the original image data of the sign is subjected to structured analysis. The Sobel operator is introduced to calculate gradients in the horizontal and vertical directions. A pre-defined convolution kernel is used to convolve the pixel grayscale changes, thereby extracting edge intensity information. The generated image edge feature data has high texture clarity and can effectively reveal the structural contours of the text edges. The original image and the edge feature image are then weighted and fused proportionally. The original image retains the overall structural information, while the edge image highlights detailed boundaries. Gaussian filtering is then applied to the fused result to obtain an edge-enhanced fused image. Adaptive threshold segmentation is applied to the edge-enhanced fused image. Local means and standard deviations are calculated for different local regions, and a segmentation threshold is set based on these. Pixels that meet the criteria are marked as text regions, effectively generating a binary image with clear foreground and background distinction. The binary text region image is input into a multi-scale text recognition network. Multiple convolution branches are set in parallel within the network, using convolution kernels of different scales to extract multi-level features of the image at the detail, small structure, and global semantic levels. This network introduces residual connections in each convolutional layer, enabling efficient flow and fusion of features between layers, improving recognition efficiency and accuracy. Finally, it outputs the character sequence to be converted via a sequence decoding mechanism. The character sequence is then subjected to semantic encoding analysis based on semantic context. A sliding window is constructed to extract the contextual phrases for each character, which are then input into a pre-trained language model to generate a semantic vector. A polyphonic character disambiguation module calculates the most likely pronunciation of the current character in a specific semantic environment. Based on this pronunciation, its unique dot code is retrieved from a pre-defined Chinese character-Braille mapping matrix. Finally, combining the dot coordinates, height, and diameter parameters of the Braille code, a Braille dot imprinting instruction set is constructed, resulting in an actuator-driven instruction set.

[0023] Step 200: Based on the actuator drive instruction group, synchronously drive the printing actuator to perform Braille dot imprinting operation, and collect imprinting force and position deviation signals in real time to obtain actuator operation status feedback data. Specifically, based on the Braille dot coordinates, imprint height, imprint diameter, and imprint force parameters contained in the actuator drive command group, the imprint head in the imprint actuator is driven to move along a set path on a two-dimensional coordinate plane, and performs an imprinting operation after reaching the target imprint position. During the process of the imprint head contacting the surface of the sign substrate, a piezoresistive force sensor configured on the imprint head is activated to sample the normal pressure generated between the imprint contact interfaces in real time. This sensor continuously acquires the instantaneous value of the imprint force at a fixed frequency and outputs a set of time-series data, reflecting the dynamic change trend of the pressure throughout the entire imprinting cycle. Simultaneously, to ensure the forming accuracy of each imprint point in space, a high-precision photoelectric encoder module configured on the printing mechanism is synchronously invoked to read the actual displacement values ​​of the current imprint head in the x-axis and y-axis directions in real time, compare them with the target coordinates, calculate the spatial position error value at the current moment, and effectively identify imprint position deviation problems caused by transmission lag, path drift, or mechanism vibration. Based on the imprinting force time series data and position error data, the spatial coordinates (including its two-dimensional planar position) of the Braille point currently being imprinted are added to the dataset as identification information through a structured annotation mechanism. At the same time, a timestamp is added to each set of data to form actuator operation status feedback data containing time series, spatial and error information.

[0024] Step 300: Input the actuator operating status feedback data into the dual-channel attention control network for dynamic parameter adjustment to obtain the first actuator control signal and the completed Braille dot matrix; It should be noted that the actuator operating status feedback data undergoes type classification and structured processing. This feedback data includes continuous pressure values ​​collected during the imprinting process and real-time position error information in the x and y axes. Therefore, during the data access phase, the raw operating status feedback data is divided into channels based on physical properties using a structured preprocessing module. The imprinting force time-series data is assigned to the pressure channel input data, while the lateral and longitudinal error information related to the actuator's spatial displacement is assigned to the position channel input data. A dual-channel attention control network PPCA-Net is activated, with independent attention calculation paths configured for the pressure and position channels. The pressure channel analyzes the fluctuation trend and deviation of the imprinting force from the standard value over different time periods to calculate the importance weight of the pressure influence in the current control cycle. The position channel generates an importance assessment weight for the position error influence based on the real-time absolute value of the deviation, the offset trend, and the instantaneous drift rate in the X and Y directions. Each channel internally executes a set of attention mechanism weight functions, outputting single-channel attention coefficients through softmax normalization. Then, the attention results of the two channels are weighted and merged according to the set fusion coefficient to obtain the fused attention control factor. This factor reflects the comprehensive influence of the two types of errors on the actuator control behavior during the current imprint point control process. Based on the fused attention control factor, a compensation control quantity is calculated to correct the basic control signal calculated by the traditional PID control model, thus forming a more responsive composite adjustment term. This compensation control quantity is then linked with the actuator dynamics model, which considers three dynamic constraints: the actuator's mass, damping coefficient, and stiffness coefficient. The calculation result is output as the first actuator control signal, which includes imprint force control dynamically adjusted based on error assessment, displacement compensation commands, and motion trajectory fine-tuning commands, synchronously driving the imprint actuator to perform the imprinting action at the current Braille dot. After the imprinting operation at the current Braille dot is completed, its position coordinates, forming status, feedback data, and quality detection information are synchronously written into the Braille dot matrix update buffer, thereby completing the real-time refresh of the overall printing status and obtaining the completed Braille dot matrix.

[0025] Step 400: Perform optical inspection and geometric dimension measurement on the completed Braille dot matrix to obtain Braille dot forming quality deviation data; Specifically, high-precision laser displacement sensors deployed at the end of the printing execution module or along the detection trajectory are used to detect the vertical height of each Braille dot after it has been imprinted. Based on the principle of non-contact reflection, the laser displacement sensor measures the time or phase difference required for the laser beam to return from the top of the Braille dot, obtaining the actual height of the dot's elevation. This constructs a dataset of Braille dot height measurements with spatial order attributes, reflecting whether each dot has reached the set standard height (e.g., 0.5mm) during the imprinting process and revealing insufficient imprinting or overshooting. Simultaneously, to understand the lateral forming characteristics of the Braille dots, a high-resolution CCD image acquisition device is used to obtain a top-down image of the Braille dot area. After image calibration and denoising, a contour boundary extraction algorithm is applied to perform edge recognition and contour modeling for each Braille dot. Then, the diameter value of each Braille dot is calculated based on the distance between the midpoint of the contour and the edge, forming a diameter measurement dataset. This diameter dataset can reveal whether there are structural problems such as edge expansion, shrinkage, or deformation of the Braille dots. Difference analysis was performed on the height and diameter measurement datasets and preset standard values, respectively. The standard height was set to 0.5 mm, and the standard diameter to 1.5 mm. The deviation between each actual measurement value and its standard value was calculated point-by-point, forming subsets for height and diameter deviations. Based on this, the two types of geometric deviation information were structurally integrated to generate a set of Braille dot geometric deviation parameters. A molding quality deviation data structure was constructed based on the spatial coordinates, timestamp, and deviation amplitude of each Braille dot. The distribution characteristics of statistical parameters such as standard deviation and mean offset during the calculation process were considered to determine whether there were batch anomalies or systematic error trends. This generated the Braille dot molding quality deviation data.

[0026] Step 500: Based on the Braille dot forming quality deviation data, perform gradient analysis and compensation on the first actuator control signal to obtain the second actuator control signal for the next printing process.

[0027] Specifically, the time derivative of key numerical terms in the Braille dot forming quality deviation data is calculated to obtain the instantaneous rate of change of Braille dot quality over time, resulting in a set of deformation gradient data describing the stability trend of the printing process. This data can reveal abnormal changes at individual points and reflect the degree of actuator state degradation over time through trend indicators. A sliding window analysis method is applied to the deformation gradient data, setting a fixed-length time window and performing step-by-step calculations on the deviation curve sequence. The mean, standard deviation, and abrupt change point identification algorithm of the gradient values ​​within the window are used to extract the trend of actuator output performance changes. If the analysis results show that the deviation gradient of the Braille dots continuously increases or fluctuates drastically within several consecutive window periods, and this trend exceeds the system's preset attenuation threshold, it is preliminarily determined that the current actuator has exhibited performance degradation behavior. The system enters a fault warning state and triggers a fault diagnosis process to generate an actuator performance compensation strategy selection instruction. Based on this instruction, the various control gain parameters contained in the first actuator control signal—including proportional gain, integral gain, and derivative gain—are adaptively adjusted. The adjustment strategy, based on the aforementioned gradient trend data and combined with the controller's historical performance and known actuator response characteristics, dynamically generates the optimal target control parameter set through mechanisms such as increasing response speed with proportional gain, compensating for system deviations with integral gain, and suppressing oscillations with derivative gain. This adapts to the control requirements after actuator performance degradation. To ensure the long-term effectiveness of the control strategy and the stability and sustainability of the printing process, an actuator lifetime prediction model is introduced. The integral value of the deformation gradient within the current control cycle and historical operating load data are input into the lifetime assessment module to predict the remaining lifetime. Based on this, the application weights of the target control parameter set are adjusted or a backup control configuration is invoked. The adjusted target control parameter set and the lifetime assessment results are then combined to generate the second actuator control signal.

[0028] In one specific embodiment, the process of performing step 100 may specifically include the following steps: Sobel gradient calculation and edge intensity extraction are performed on the text image of the sign to obtain image edge feature data; Weighted fusion and Gaussian filtering are performed on the image edge feature data and the sign text image to obtain an edge-enhanced fused image; Adaptive threshold segmentation is performed on the edge-enhanced fused image to obtain a binarized text region image; The binarized text region image is input into a multi-scale recognition network for residual connection feature extraction and sequence decoding to obtain the text character sequence to be converted. Semantic context analysis and Braille encoding mapping are performed on the character sequence to be converted to obtain the actuator drive instruction set.

[0029] Specifically, the text image on the original signboard undergoes preprocessing. The Sobel operator is used to calculate gradients along both the horizontal and vertical directions of the image. This process extracts pixel regions with the most dramatic gray-level changes at the edges by using a specific convolution kernel to perform differential calculations on the image's gray-level distribution. These regions correspond to text outlines, edges, and boundaries. The gradient map generated through this process enhances the areas with significant structural changes in the image and provides boundary information. The image edge feature data and the signboard text image are then weighted and fused. The fusion ratio is balanced by adjusting the weight coefficients of the original image and the edge map, for example, using a combination of α = 0.7 and β = 0.3. After the fusion operation, to reduce noise interference in the image and smooth pixel abrupt changes at the fusion boundary, the system uses a Gaussian filtering algorithm to smooth the fused image. This operation effectively suppresses the influence of high-frequency noise by weighted averaging of pixel values ​​within local regions, resulting in more continuous and clear character edges. After filtering, the system performs adaptive thresholding on the fused image. It automatically calculates the most suitable threshold based on the mean and standard deviation of the grayscale values ​​of each local region in the image, determining whether each pixel belongs to the foreground or background, and generating a binary image. White areas represent the identified text, while black areas represent the background. The system inputs the binary image into a multi-scale recognition network. This network has three parallel convolutional paths responsible for extracting detailed features, small-scale structures, and global contextual information, respectively. A residual connection mechanism is introduced between the convolutional layers to maintain stable gradient flow during training, effectively preventing gradient vanishing or overfitting. Simultaneously, each branch is equipped with a channel attention mechanism, allocating different levels of attention among the output channels of each convolution, allowing the network to focus on the most discriminative structural regions in the image. After convolutional extraction, attention weighting, and fusion operations, the network performs time-series parsing of the text information in the feature map through a sequence decoding layer. A temporal classification algorithm is then used to reconstruct the character sequence, resulting in the text character sequence to be converted. Semantic context analysis is then performed on the text character sequence to be converted. The system constructs a context window for each target character, consisting of several neighboring characters before and after it. This context sequence is then input into the pre-trained language model BERT, which encodes the semantic representation of the current character within that context using its deep multi-head attention structure. Based on this semantic representation, the system predicts the pronunciation of Chinese characters with multiple pronunciations using a disambiguation module. The most suitable Braille code is selected based on the predicted phonetic semantic category. The system extracts the corresponding dot code from a pre-constructed Chinese character-Braille dot matrix using a lookup table method. This code includes the dot matrix distribution corresponding to the character, as well as the spatial coordinate information of the standard Braille structure.After converting all characters to Braille dots, the system generates molding control parameters for each Braille dot according to its geometric definition, including its X and Y coordinates, raised height (standard 0.5mm), and dot diameter (standard 1.5mm). Finally, all dot control commands are summarized into an actuator drive command group, which expresses the complete printing path, imprinting instructions, and sequence control information in structured data form.

[0030] In one specific embodiment, the process of inputting the binarized text region image into a multi-scale recognition network for residual connection feature extraction and sequence decoding to obtain the text character sequence to be converted can specifically include the following steps: The binarized text region image is simultaneously distributed to three parallel feature extraction branches in the multi-scale recognition network. The small-scale branch of the three parallel feature extraction branches is used to capture the text detail edge features, the medium-scale branch of the three parallel feature extraction branches is used to extract the local structural features of the text, and the large-scale branch of the three parallel feature extraction branches is used to obtain the global features of the text, resulting in three sets of multi-scale original feature maps with different receptive fields. Deep feature learning and channel concatenation are performed on three sets of original feature maps with different receptive fields at multiple scales to obtain multi-scale fused feature maps. The feature vectors of the multi-scale fused feature maps are then serialized and rearranged to obtain feature sequences. The feature sequence is input into the decoder for forward computation, while the merging rules remove duplicate characters and whitespace markers to obtain the text character sequence to be converted.

[0031] Specifically, the binarized text region image is simultaneously distributed to three parallel feature extraction branches in the multi-scale recognition network. Each branch uses convolutional kernels of different sizes to simulate the multi-layered visual perception process of the human eye when observing text, from details to the whole. The first small-scale branch primarily uses 3×3 convolutional kernels, with a smaller receptive field, suitable for capturing high-frequency areas in the image, such as stroke edges, corners, breakpoints, and subtle structural changes. This is suitable for recognizing blurry or tightly packed text images. In this path, the system uses multi-layer convolution combined with the ReLU activation function, and superimposed batch normalization structures to improve feature stability. This allows the convolution results to accurately characterize the contour changes and boundary closure characteristics of characters, and uses residual connection structures to avoid feature degradation, ensuring that image detail features are fully preserved in the deep structure. Meanwhile, the second medium-scale branch uses a 5×5 convolutional kernel structure, expanding the receptive area to identify local structural relationships in the text, such as radicals, top-bottom structures, left-right symmetry, and character frames. These structures carry the word-formation characteristics in Chinese character recognition and are key clues for distinguishing similar-shaped Chinese characters. In this path, convolutional features enhance the expressive power from stroke combinations to structural configurations through progressive integration between layers, while the residual structure ensures no information loss. The third large-scale branch uses 7×7 or larger convolutional kernels to capture the contextual semantic structure of the entire character and its neighboring characters. This helps in context judgment and character splitting when there is character adhesion, dense layout, or occlusion. This branch can model the spatial distribution of the entire line of text and simulate the syntactic layout characteristics in text arrangement. The three branches output three sets of original feature maps, each with its own emphasis on spatial receptive field and representational ability, possessing a complete expressive capability from details to the whole. These three sets of feature maps are input into the deep feature learning module, where a dimension regularization operation is performed to ensure channel compatibility. Then, a channel concatenation operation is performed to merge the three sets of feature maps along the channel dimension, forming a unified multi-scale fused feature map. During the fusion process, the system introduces an attention mechanism to assign higher weights to important feature channels and suppress redundant channel noise, thereby optimizing feature quality. The fused feature map is then input into the sequence generation module. The system performs serialization and rearrangement on the fused feature maps, converting the two-dimensional feature maps into a sequence of one-dimensional feature vectors along the column or row direction while preserving the spatial order of the original images. This serialization operation combines techniques such as global average pooling, feature map unrolling, and positional encoding to ensure that each vector in the sequence contains both local features of a specific location in the image and retains its relative position information within the entire image. The resulting feature sequence is an ordered set of vectors, each representing a composite feature representation of a character region in the text image. The feature sequence is then input into a decoder for character recognition. The decoder employs a forward computation mechanism, independent of fixed segmentation points, and recognizes the entire line of text through end-to-end prediction.During the decoding process, the system uses the CTC (Connected Temporal Classification) method to process the sequence, allowing the output to contain whitespace markers and repeated characters, and without mandatory alignment, thus enabling the training of a high-precision recognition model even without character boundary labels. The CTC layer performs a redundancy removal operation at the output, first merging consecutively repeating character labels, and then removing whitespace markers, thereby generating a clean, non-repeating, and accurate text sequence. In a specific embodiment, the process of performing semantic context analysis and Braille encoding mapping on the text character sequence to be converted to obtain the actuator-driven instruction set can specifically include the following steps: After extracting contextual information from each target character in the character sequence to be converted, the input is fed into the BERT pre-trained model for context semantic vector encoding to obtain the character context representation vector. Input the character context representation vector into the polyphonic character disambiguation module, determine the unique phonetic annotation from multiple candidate pronunciations of the polyphonic character disambiguation module, and obtain the disambiguated character pronunciation sequence; Based on the disambiguated character pronunciation sequence, the pre-constructed Chinese character-Braille mapping matrix is ​​searched to obtain the corresponding six-dot Braille code. Then, each six-dot Braille code is converted into Braille dot matrix spatial distribution data in a two-dimensional spatial coordinate system through the Braille dot coordinate transformation function. Based on the spatial distribution data of Braille dots, an actuator drive instruction set containing dot coordinates, imprint height, imprint diameter, and imprint force is generated.

[0032] Specifically, a local context window containing background information is constructed for each target character. The window length is set to an odd number (e.g., 7) to ensure the current character is centered. A corresponding number of neighboring characters are extracted from its left and right sides to form a semantically continuous character segment. This character segment is input into a pre-trained language model based on the BERT architecture for contextual semantic vector encoding. The BERT model employs a multi-layer bidirectional Transformer structure, which can simultaneously model the influence of left and right word meanings on the target character without introducing a fixed directional bias, thereby generating a high-dimensional semantic representation vector with context awareness. This representation vector contains information about the character itself and reflects its semantic tendency and contextual association features in the current language environment. In this way, each character is assigned a semantic vector, which can effectively distinguish the actual pronunciation meaning of polyphonic characters in different contexts. The system inputs the above character context representation vector into a polyphonic character disambiguation module. This module contains several speech category discriminators. Each candidate pronunciation is treated as a category, and a decision boundary is constructed in the vector space. The probability value of each candidate pronunciation is calculated using a softmax activation function, forming a set of pronunciation candidate distributions. In this distribution, the system selects the pronunciation with the highest confidence as the unique phonetic annotation of the character in the current context according to the principle of maximum probability, thereby achieving accurate disambiguation of polyphonic characters. A pre-constructed Chinese character-Braille mapping matrix is ​​then searched based on the disambiguated character pronunciation sequence. Each character and its corresponding unique pronunciation are used as index entries, and a search operation is performed in the Chinese character-Braille mapping matrix. This mapping matrix uses the character's phonetic representation as the primary key, with each entry corresponding to a standard six-dot Braille code. This code is a 6-bit binary sequence representing the combination of dots that should be raised in the six-dot Braille structure. After extracting the six-dot Braille codes, the system converts these codes into geometric spatial data recognizable by the actuator. It calls the Braille dot coordinate transformation function to map the corresponding raised dot in each code to a specific coordinate position in two-dimensional space. Using a three-column, two-row arrangement, the X-coordinate is defined as being in the interval [0,2] and the Y-coordinate as being in the interval [0,3]. The physical position of each blind dot in the local Braille unit is calculated uniformly according to the dot numbering rules. To achieve a complete Braille character layout, the system cascades and stitches together all individual Braille units in a planar manner, based on parameters such as Braille character spacing, row and column arrangement rules, and text orientation, to form a Braille dot matrix spatial distribution map. This map contains the two-dimensional spatial coordinates of each blind spot and also implicitly includes the character arrangement relationships and structural spacing. Based on this, the system combines the spatial coordinates of each blind spot with imprinting-related physical parameters to form a dot control unit data structure.The imprint height is 0.5mm to ensure tactile perceptibility of the Braille, and the standard imprint diameter is 1.5mm to meet national standards for the visual and tactile adaptability of tactile dot structures. The imprint pressure is adaptively calculated by the control system based on material hardness and actuator characteristics, or set according to default values, ensuring the dots are fully formed without penetrating the substrate. The system packages the above information into actuator drive command sets. Each command includes the X and Y coordinates of the tactile dot on the printing panel, the required imprint height and diameter, and the corresponding pressure level. All commands are sorted according to character structure and dot order to form a data sequence.

[0033] In one specific embodiment, the process of performing step 200 may specifically include the following steps: The printer drives the printing head to move to the target position and perform Braille dot imprinting operation according to the point coordinates, imprint height, imprint diameter and imprint force in the actuator drive instruction group; During the ongoing Braille dot formation process, the real-time imprinting force between the imprint head and the substrate is continuously sampled and monitored using a piezoresistive sensor to obtain imprinting force time-series data. Simultaneously, photoelectric encoders are used to measure and calculate the actual position of the printing actuator in the x-axis and y-axis directions in real time to obtain position error data; Add the spatial coordinates and time markers of the current imprint point to the imprint force timing data and position error data to obtain actuator operating status feedback data.

[0034] Specifically, the system receives a structured set of drive instructions from the encoding generation module. Each instruction corresponds to a Braille dot to be printed. This instruction includes the dot's position parameters in a two-dimensional coordinate system, the desired imprint height, the standard diameter, and the appropriate imprint pressure level. Upon receiving this set of instructions, the actuator control module issues a path scheduling command to the mechanical platform or multi-axis drive system based on the coordinate information of each instruction. A high-speed, high-precision displacement control system quickly guides the imprint head to the target position. After confirming stable positioning, the system initiates an imprint command, driving the imprint head to apply pressure to the substrate surface along the normal direction to complete the local deformation and forming of the Braille dot. This action is controlled by the collaborative mechanism of the position control system and the pressure application module. Pressure control is initiated after the movement is in place, ensuring the imprint height is controlled within a set value (e.g., 0.5mm) plus or minus 0.1mm, and the diameter is maintained at approximately 1.5mm. The imprinting process is stably executed under timing scheduling and execution redundancy protection. During the imprinting process of the Braille dot, the system is not controlled in an open-loop manner but rather a closed-loop feedback path is activated in real time. The piezoresistive sensor integrated within the printing module monitors the contact pressure between the impression head and the substrate throughout the entire printing process. This sensor continuously samples the pressure value at every moment during the printing process with a millisecond sampling period, generating a time-series curve of the impression force. This curve records the upward trend during the pressure application phase and includes the complete physical changes during the maximum pressure point, the stabilization phase, and the depressurization phase. This allows the system to reflect typical instability phenomena such as weak pressure, excessive impact, or insufficient rebound during printing. Simultaneously, to ensure absolute accuracy in spatial positioning, the system uses a high-resolution photoelectric encoder mounted on the platform to continuously measure the actual movement trajectory of the impression head in the X and Y axes. The photoelectric encoder converts the real-time displacement signal of the impression head into position values ​​and compares them with the coordinates of the points set in the target command, thereby calculating the lateral and longitudinal position errors at the current moment. This position error reflects the degree of deviation between the actuator's theoretical trajectory and the actual trajectory, and indirectly reveals spatial control errors caused by inertial response, mechanism backlash, or motion disturbances. The system uses this error information to correct the target path for subsequent points, improving overall printing consistency. The error value is recorded as a sampling sequence, allowing each imprint point to generate an independent dynamic position error curve. Spatial coordinates of the current imprint point are added to the imprint force timing data and position error data. These spatial coordinates are derived from the original coordinate information in the actuator drive command group and are used to bind the feedback data to the actual control target. Simultaneously, to achieve time-series tracking and fault location analysis, the system timestamps each data set, recording the start and end times of data acquisition and the physical occurrence time of the imprint operation. Through these steps, actuator operating status feedback data is obtained.

[0035] In one specific embodiment, the process of performing step 300 may specifically include the following steps: The actuator operating status feedback data is separated into pressure channel input data and position channel input data according to the data type. The pressure channel input data includes imprint force timing data, and the position channel input data includes position error data in the x and y directions. The pressure channel input data and the position channel input data are respectively input into the dual-channel separate control network to calculate the attention weights and obtain the dual-channel attention weight coefficients. The attention control factor is obtained by weighted fusion based on the dual-channel attention weight coefficients. The compensation control quantity is calculated based on the fused attention control factor, and the first actuator control signal is generated based on the actuator dynamics model and the compensation control quantity. The printing actuator is driven by the control signal of the first actuator to complete the imprinting and forming operation of the current Braille dot and update the overall printing status to obtain the completed Braille dot matrix.

[0036] Specifically, the system performs data type separation on the collected feedback data. Based on data attributes and physical meaning, the system categorizes data into channels, using continuous data recording the contact pressure change curve during the imprinting process as input data for the pressure channel, while spatial position deviations in the X and Y directions are uniformly classified as input data for the position channel. The system inputs the data from both channels into a structurally symmetrical but functionally independent dual-channel separation control subnetwork. This subnetwork features a parallel attention mechanism, enabling it to independently and dynamically model the importance of information within each channel. In the pressure channel, the network standardizes and embeds the imprinting force time-series data, mapping the pressure value at each moment into a high-dimensional vector before inputting it into a multi-head attention layer. The built-in weight matrix is ​​used to calculate the influence of pressure at each moment on the overall imprinting process, thereby generating pressure attention weight coefficients describing the weight distribution of each data point in that channel. Simultaneously, the network in the position channel normalizes the X-axis and Y-axis error data and merges them into a two-dimensional error vector. This vector is then input into the position attention module via feature embedding. In this module, the system jointly models the error amplitude, error change rate, and position offset direction. An adaptive attention distribution function assigns response weights to each position error information, outputting position attention weight coefficients. The system weights the two sets of attention weights according to the fusion principle set in the control strategy, forming a fused attention control factor. This fusion process is based on an empirical fusion coefficient λ, ranging from 0.5 to 0.7, used to balance the priorities of force control and position control in the current printing task. When the deformation accuracy requirement is high in the printing scenario, the weight of the pressure channel is increased; conversely, when the priority of positioning accuracy is high, the proportion of the position channel is increased accordingly. A compensation control vector is calculated based on the fused attention control factor. This vector serves as a gain adjustment factor applied to the PID control structure to compensate for deviations in the basic control model caused by equipment aging, external disturbances, or feedback lag. Simultaneously, the system invokes the actuator dynamics model as a constraint condition. This model describes the actuator's mechanical characteristics, such as mass distribution, damping coefficient, and elastic stiffness. By simultaneously calculating the compensation control quantity with the dynamics model, a first actuator control signal that meets the dynamic response requirements and is physically feasible is generated. This signal includes path instructions and target position, as well as dynamically adjusted pressure application timing, acceleration boundaries, and velocity transition zone control strategies. This first control signal is sent to the actuator control interface to drive the imprinting mechanism to perform the imprinting task of the current Braille point. During the movement of the imprint head to the target point, the system controls the actuator to accurately position itself using pre-loaded path planning data and inertial compensation parameters. After stable positioning, the pressure application module is activated to perform force control operations. Through the above control signals, the imprint head will complete the blind spot formation at the set height and target diameter. During this process, the system continuously collects imprinting force and position feedback data to synchronously verify the actual control effect.Once the Braille dot imprinting task is completed, the system marks the dot as "completed" and writes its spatial coordinates, imprinting parameters, feedback error, and quality inspection information into the Braille dot matrix status cache, updating the overall printing status diagram.

[0037] In one specific embodiment, the process of calculating the compensation control quantity based on the fused attention control factor and generating the first actuator control signal based on the actuator dynamics model and the compensation control quantity can specifically include the following steps: By performing a difference calculation between the target position coordinates of the current imprint point and the actual measured position coordinates, and simultaneously performing a difference calculation between the target imprint force and the actual imprint force, a current tracking error vector containing position tracking error and pressure tracking error is obtained. Based on the current tracking error vector, the proportional term, integral term and derivative term are calculated using the PID control law to obtain the basic PID control quantity; The fused attention control factor is multiplied by the preset compensation gain coefficient to generate the compensation control quantity; The actuator dynamics model established based on the actuator mass matrix, damping matrix and stiffness matrix is ​​used to perform dynamic constraint verification on the PID basic control quantity and compensation control quantity, and the adjustment control quantity under dynamic constraints is obtained. The control signal is amplitude modulated and timing adjusted based on the adjustment control quantity under dynamic constraints and the current actuator state information to obtain the first actuator control signal.

[0038] Specifically, during each imprinting process, the system extracts the spatial coordinates of the current target imprinting point from the actuator drive command set, including the X and Y position parameters in the two-dimensional plane, as well as the set target imprinting force value. Simultaneously, it calls upon the current position data and current imprinting force measurement values ​​provided in real-time by the photoelectric encoder and piezoresistive sensor in the feedback acquisition system. By calculating the difference between the target position coordinates and the actual measured position coordinates, and by calculating the difference between the target imprinting force and the actual imprinting force, a set of three-dimensional tracking error data is obtained. This includes the x-axis position error, y-axis position error, and imprinting force error. These three errors together constitute the current tracking error vector, characterizing the degree of deviation of the current actuator state from the theoretical control target. The tracking error vector is then input into the PID controller for control law expansion. The controller internally calculates a proportional term based on each error, directly multiplying the error value by a set proportional gain coefficient to reflect the immediate response capability of the control system to the current deviation. The system calculates an integral term by weighted integration of the accumulated error over time to compensate for long-term stable deviations and enhance the system's steady-state accuracy. The system calculates the derivative of the error over time, i.e., the rate of change of error, and combines it with a differential gain coefficient to form a differential term, preventing system oscillations or response overshoot caused by excessively rapid deviation changes. These three control components together constitute the basic PID control quantity. The fused attention control factor is multiplied by a preset compensation gain coefficient to form a compensated control quantity. This compensated control quantity adaptively adjusts in time, amplitude, and direction according to the attention mechanism, effectively extending the error domain that traditional controllers cannot cover. It is suitable for maintaining imprinting accuracy under complex conditions such as material hardness fluctuations, platform inertia changes, or actuator performance degradation. To ensure that the compensated control quantity does not exceed the actual limitations of the actuator's physical structure or dynamic response, the system constructs an actuator dynamic model and uses this model as a basis for physical constraint verification of the control signal. This dynamic model, with mass, damping, and stiffness matrices as its core structure, describes the actual motion behavior and response inertia of the actuator under a unit control signal. The system inputs the superimposed PID basic control quantity and compensation control quantity into the model equations, and calculates through dynamic simulation whether the current control quantity will induce unstable states such as system resonance, overload, or structural response delay. If it is found that the control signal will exceed the maximum load-bearing capacity or cause structural impact under the current state, the system will weaken or dynamically correct the control quantity based on model feedback, outputting an adjustment control quantity under dynamic constraints to ensure that the control signal meets the control objective without violating physical boundary conditions, thus guaranteeing the long-term stable operation of the actuator structure. Based on the current operating status information of the actuator, including current position, velocity, cumulative load, and environmental disturbance parameters, the system performs amplitude modulation and timing adjustment on the adjustment control quantity under the aforementioned dynamic constraints.During amplitude modulation, the system considers the current actuator's driving linear region, energy consumption threshold, and hardware limitations, compressing or expanding the signal amplitude. In multi-cycle control, it executes an amplitude limiting protection strategy to prevent the controller from outputting excessive excitation in a short period of time, which could damage the mechanism. In the timing adjustment phase, the system adjusts the rising edge of the control signal, platform response lag compensation, and signal smoothing interpolation to ensure that the control quantity is precisely matched with the actuator's response capability in the time dimension. This avoids deviations between the actual execution path and the expected trajectory due to uneven signal time distribution, ultimately obtaining the first actuator control signal.

[0039] In one specific embodiment, the process of performing step 400 may specifically include the following steps: A laser displacement sensor was used to optically detect the height of each Braille dot in the completed Braille dot matrix, resulting in a Braille dot height measurement dataset. Image acquisition and contour boundary extraction are performed on the completed Braille dot matrix, and the Braille dot diameter measurement dataset is calculated; The Braille dot height measurement dataset and the Braille dot diameter measurement dataset are respectively compared with the standard height value and the standard diameter value to obtain the geometric deviation parameter set, and the Braille dot forming quality deviation data are generated based on the geometric deviation parameter set.

[0040] Specifically, to obtain the actual height information of each Braille dot in the resulting three-dimensional structure after molding, the system uses a laser displacement sensor to scan the Braille dot matrix point by point. This laser displacement sensor utilizes the principle of reflective non-contact ranging, emitting a laser beam to the top of the Braille dot and receiving the reflected signal. It calculates the reflection time or phase shift, and then converts this into the vertical displacement value of the point, thus obtaining the actual height data relative to the substrate reference plane. In actual execution, the laser sensor scans and positions itself in a dot matrix format, systematically traversing the blind spot area according to the Braille character structure and dot matrix arrangement rules, ensuring that each independent dot is completely scanned and a unique height value is generated. After scanning, the system numbers and archives all collected height values, constructing a Braille dot height measurement dataset according to the order of the Braille dots. This dataset reflects the height distribution of individual dots and reveals whether there are significant height differences between adjacent dots and batch molding consistency issues. Simultaneously, to evaluate the horizontal dimensional molding of the Braille dots, the system also performs dot diameter measurements. This task relies on an image acquisition system built with a high-definition CCD camera and an image processing module. This system captures a top-down image after the Braille dot matrix is ​​formed, and improves image clarity through preprocessing techniques such as distortion correction, contrast enhancement, and image denoising to ensure stable and discernible edge features. After image acquisition, the system performs contour boundary extraction on the image data. This process uses edge detection algorithms, such as the Canny or Sobel operators, combined with contour tracking technology to identify the closed boundary of each Braille dot in the image, thereby extracting the outer contour of the blind spot. The system mathematically models each boundary using contour fitting techniques (such as ellipse fitting or minimum circumcircle fitting), and calculates the actual diameter of each Braille dot based on the image scaling ratio and the actual physical size ratio, forming a Braille dot diameter measurement dataset. This dataset reflects whether each blind spot exhibits lateral structural defects such as edge expansion, collapse, or edge contraction. The Braille dot height measurement dataset and the Braille dot diameter measurement dataset are then compared with standard height and diameter values, respectively, using difference calculations. According to the national standard GB / T 15720, the standard height of a Braille dot is set to 0.5 mm, and the standard diameter is 1.5 mm. Therefore, the system subtracts 0.5 mm from each value in the height measurement dataset to obtain the height deviation value for each dot. Similarly, it subtracts 1.5 mm from the diameter measurement dataset to obtain the corresponding diameter deviation value. These two sets of deviation data are structured and integrated into a geometric deviation parameter set. Each data item in the parameter set records the spatial coordinates, height deviation, and diameter deviation of a single Braille dot, and introduces descriptive parameters such as standard deviation, maximum offset, and average deviation within the region to characterize the uniformity and stability of the overall printing batch's forming quality. Braille dot forming quality deviation data is generated based on the geometric deviation parameter set.

[0041] In one specific embodiment, the process of performing step 500 may specifically include the following steps: The time derivative of the Braille dot forming quality deviation data is calculated to obtain the deformation gradient data. Based on the deformation gradient data, a sliding window analysis is performed to obtain the result of the actuator performance degradation trend judgment. When the result of the actuator performance degradation trend judgment exceeds the preset degradation threshold range, the fault diagnosis process is initiated and an actuator performance compensation strategy selection instruction is generated. Based on the actuator performance compensation strategy, the control gain parameter in the control signal of the first actuator is adaptively adjusted according to the selected instruction to obtain the target control parameter set; The target control parameter set is combined with the actuator remaining life prediction data to generate the second actuator control signal for the next printing process.

[0042] Specifically, the system calculates the time derivative of the Braille dot forming quality deviation data and performs differential calculations on the quality deviation changes between two or more consecutive printing cycles to obtain deformation gradient data representing the rate of change in the forming state of each dot or region. Based on this deformation gradient data, the system performs sliding window analysis. A fixed time window length and sliding step size are set, and within this window, the system statistically analyzes characteristic indicators such as the mean, maximum, and standard deviation of the gradient data, constructing a local trend map of actuator performance changes. If, within multiple consecutive sliding window cycles, the gradient value exhibits a monotonically increasing trend, an increased fluctuation range, or a long-term deviation from the normal fluctuation range, the system determines that the actuator is experiencing abnormal trends such as decreased output power, changes in mechanism damping, control response delay, or attenuation of force-position coupling efficiency during execution. This generates a set of performance trend judgment results characterizing actuator state degradation. These judgment results are compared with a preset attenuation threshold. When the trend indicator in any dimension exceeds this threshold range, it indicates that the actuator has deviated from its normal operating state and entered a potential degradation region. When the system detects a performance degradation trend, it initiates a fault diagnosis process. Based on the current degradation mode, the system matches the most likely failure mechanism from a predefined performance degradation type library, such as control hysteresis, energy decay, or structural friction, and generates an actuator performance compensation strategy selection instruction accordingly. This instruction specifies the type of compensation mechanism and includes scheduling information such as the required control parameter correction range, adjustment target priority, and compensation execution time scale. The controller adaptively adjusts the control gain parameters in the original first actuator control signal according to the strategy category in the instruction, including numerical updates of proportional gain, integral gain, and derivative gain. In specific execution, the system updates the PID gain coefficients using linear, exponential, or step gain adjustment functions based on the magnitude and duration of the deformation gradient slope, enhancing feedback response capability and improving the system's steady-state correction capability for persistent deviations, thereby forming a set of target control parameters that match the current actuator's actual state. Based on parameters such as control signal strength, displacement load, printing cycle, and temperature changes from past printing tasks, an actuator remaining life prediction model is constructed with cumulative load integral and degradation rate factor as inputs to calculate the current actuator's life stage and remaining usable cycles. When the prediction results show that the remaining lifespan is below the warning threshold (e.g., 20%), the system activates a workload reduction mechanism. This involves adjusting the current target control parameter set by reducing the control gain or increasing the control buffer while still meeting printing accuracy requirements. This reduces the actuator's operating intensity, extending its physical lifespan and controlling long-term wear. The system then jointly encodes the adjusted target control parameter set with the actuator lifespan prediction data. Based on the current task structure and the complexity of the Braille dot matrix, it generates the second actuator control signal for the next printing cycle through an optimized scheduling strategy.The control signal includes PID control quantities adjusted by dynamic compensation and lifespan constraints, as well as additional control logic such as path buffer adjustment information, pressure rhythm adjustment instructions, and actuator protection threshold settings.

[0043] The printing control method in the embodiments of the present invention has been described above. The printing control system in the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 2 One embodiment of the printing control system in this invention includes: The preprocessing module is used to preprocess the text image of the sign to obtain the character sequence of the text to be converted, and to generate an actuator drive instruction group based on the character sequence of the text to be converted. The status feedback module is used to synchronously drive the printing actuator to perform Braille dot imprinting operation based on the actuator drive instruction group, and collect imprinting force and position deviation signals in real time to obtain actuator operation status feedback data. The parameter adjustment module is used to input the actuator operating status feedback data into the dual-channel attention control network for dynamic parameter adjustment, so as to obtain the first actuator control signal and the completed Braille dot matrix; The measurement module is used to perform optical inspection and geometric dimension measurement on the completed Braille dot matrix to obtain Braille dot forming quality deviation data; The compensation module is used to perform gradient analysis and compensation on the first actuator control signal based on the Braille dot forming quality deviation data, so as to obtain the second actuator control signal for the next printing process.

[0044] Through the collaborative efforts of the aforementioned components, and by employing Sobel operator edge enhancement preprocessing technology combined with a multi-scale recognition network, the challenges of recognizing complex backgrounds and blurred fonts can be effectively addressed. Compared to existing single-scale recognition methods, this significantly improves the accuracy and stability of signage text recognition under various lighting conditions and background interference. The semantic context analysis and polyphonic character disambiguation module based on the BERT pre-trained model accurately understands the correct pronunciation of Chinese characters in a preset context, effectively avoiding Braille conversion errors caused by traditional simple character mapping methods and ensuring the semantic accuracy of Braille encoding. The dual-channel attention control network, by synchronously processing pressure and position signals, achieves precise perception and dynamic adjustment of the actuator's operating state. Compared to existing open-loop control methods, it better adapts to printing requirements under different material and process conditions, ensuring the consistency of Braille dot geometric parameters. Through joint detection by laser displacement sensors and vision sensors, a complete Braille dot forming quality evaluation system is established, capable of real-time monitoring of the height and diameter parameters of each Braille dot, timely detection of quality deviations, and triggering control parameter adjustments, achieving closed-loop management of quality control. The fault diagnosis algorithm based on Braille dot deformation gradient can predict the performance degradation trend of actuators in advance. By analyzing the temporal gradient characteristics of quality changes, it realizes the transformation from passive maintenance to proactive predictive maintenance, extending equipment life and reducing maintenance costs. This invention achieves full automation of the Braille sign production process, reduces manual intervention, and improves production efficiency and product quality consistency.

[0045] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0046] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0047] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A printing control method, characterized in that, include: The text image of the sign is preprocessed to obtain a sequence of characters to be converted, and an actuator drive instruction set is generated based on the sequence of characters to be converted; Based on the actuator drive instruction group, the printing actuator is synchronously driven to perform Braille dot imprinting operation, while the imprinting force and position deviation signals are collected in real time to obtain actuator operating status feedback data. The actuator operating status feedback data is input into a dual-channel attention control network for dynamic parameter adjustment to obtain the first actuator control signal and the completed Braille dot matrix; Optical inspection and geometric dimension measurement are performed on the completed Braille dot matrix to obtain Braille dot forming quality deviation data; Based on the Braille dot forming quality deviation data, gradient analysis and compensation are performed on the first actuator control signal to obtain the second actuator control signal for the next printing process.

2. The printing control method according to claim 1, characterized in that, The process of preprocessing the signboard text image to obtain a sequence of characters to be converted, and generating an actuator drive instruction set based on the sequence of characters to be converted, includes: Sobel gradient calculation and edge intensity extraction are performed on the text image of the sign to obtain image edge feature data; The edge feature data of the image and the text image of the sign are weighted and fused and Gaussian filtered to obtain an edge-enhanced fused image; The edge-enhanced fused image is subjected to adaptive threshold segmentation to obtain a binarized text region image; The binarized text region image is input into a multi-scale recognition network for residual connection feature extraction and sequence decoding to obtain the text character sequence to be converted. Semantic context analysis and Braille encoding mapping are performed on the character sequence to be converted to obtain the actuator drive instruction set.

3. The printing control method according to claim 2, characterized in that, The step of inputting the binarized text region image into a multi-scale recognition network for residual connection feature extraction and sequence decoding to obtain the text character sequence to be converted includes: The binarized text region image is simultaneously distributed to three parallel feature extraction branches in the multi-scale recognition network. The small-scale branch of the three parallel feature extraction branches is used to capture text detail edge features, the medium-scale branch of the three parallel feature extraction branches is used to extract text local structural features, and the large-scale branch of the three parallel feature extraction branches is used to obtain text global features, resulting in three sets of multi-scale original feature maps with different receptive fields. Deep feature learning and channel concatenation are performed on the three sets of original feature maps with different receptive fields at multiple scales to obtain a multi-scale fused feature map. The feature vectors of the multi-scale fused feature map are then serialized and rearranged to obtain a feature sequence. The feature sequence is input into the decoder for forward computation, and the merging rules remove duplicate characters and whitespace markers to obtain the text character sequence to be converted.

4. The printing control method according to claim 3, characterized in that, The semantic context analysis and Braille encoding mapping of the character sequence to be converted yields an actuator-driven instruction set, including: After extracting context information for each target character in the text character sequence to be converted, the information is input into the BERT pre-trained model for context semantic vector encoding to obtain the character context representation vector. The character context representation vector is input into the polyphonic character disambiguation module, and a unique phonetic annotation is determined from multiple candidate pronunciations of the polyphonic character disambiguation module to obtain the disambiguated character pronunciation sequence. Based on the disambiguated character pronunciation sequence, a pre-constructed Chinese character-Braille mapping matrix is ​​searched to obtain the corresponding six-dot Braille code. Then, each six-dot Braille code is converted into Braille dot matrix spatial distribution data in a two-dimensional spatial coordinate system through the Braille dot coordinate transformation function. Based on the spatial distribution data of the Braille dot matrix, an actuator drive instruction set containing dot coordinates, imprint height, imprint diameter, and imprint force is generated.

5. The printing control method according to claim 1, characterized in that, The process involves synchronously driving the printing actuator based on the actuator drive command group to perform Braille dot imprinting operations while simultaneously acquiring imprinting force and position deviation signals in real time to obtain actuator operating status feedback data, including: According to the point coordinates, imprint height, imprint diameter, and imprint force in the actuator drive instruction group, the printing actuator's imprint head is driven to move to the target position and perform Braille dot imprinting operation; During the ongoing Braille dot formation process, the real-time imprinting force between the imprint head and the substrate is continuously sampled and monitored using a piezoresistive sensor to obtain imprinting force time-series data. Simultaneously, a photoelectric encoder is used to measure and calculate the actual position of the printing actuator in the x-axis and y-axis directions in real time to obtain position error data; By adding the spatial coordinates and time markers of the current imprint point to the imprint force timing data and the position error data, the actuator operating status feedback data is obtained.

6. The printing control method according to claim 1, characterized in that, The step of inputting the actuator operating status feedback data into a dual-channel attention control network for dynamic parameter adjustment to obtain the first actuator control signal and the completed Braille dot matrix includes: The actuator operating status feedback data is separated into pressure channel input data and position channel input data according to data type. The pressure channel input data includes imprint force timing data, and the position channel input data includes position error data in the x and y directions. The pressure channel input data and the position channel input data are respectively input into a dual-channel separate control network to calculate the attention weights and obtain the dual-channel attention weight coefficients. The attention control factor is obtained by weighting and fusing the dual-channel attention weight coefficients. The compensation control quantity is calculated based on the fused attention control factor, and a first actuator control signal is generated based on the actuator dynamics model and the compensation control quantity. Based on the control signal of the first actuator, the printing actuator is driven to complete the imprinting and forming operation of the current Braille dot and update the overall printing status to obtain the completed Braille dot matrix.

7. The printing control method according to claim 6, characterized in that, The step of calculating the compensation control quantity based on the fused attention control factor, and simultaneously generating a first actuator control signal based on the actuator dynamics model and the compensation control quantity, includes: By performing a difference calculation between the target position coordinates of the current imprint point and the actual measured position coordinates, and simultaneously performing a difference calculation between the target imprint force and the actual imprint force, a current tracking error vector containing position tracking error and pressure tracking error is obtained. Based on the current tracking error vector, the proportional term, integral term and derivative term are calculated using the PID control law to obtain the basic PID control quantity; The fused attention control factor is multiplied by a preset compensation gain coefficient to generate a compensation control quantity; The actuator dynamics model established based on the actuator mass matrix, damping matrix, and stiffness matrix is ​​used to perform dynamic constraint verification on the PID basic control quantity and the compensation control quantity to obtain the adjustment control quantity under dynamic constraints. Based on the adjustment control quantity under the aforementioned dynamic constraints and the current actuator state information, the control signal is amplitude modulated and timing adjusted to obtain the first actuator control signal.

8. The printing control method according to claim 1, characterized in that, The process of performing optical inspection and geometric dimension measurement on the completed Braille dot matrix to obtain Braille dot forming quality deviation data includes: A laser displacement sensor is used to optically detect the convex height of each Braille dot in the completed Braille dot matrix, and a Braille dot height measurement dataset is obtained. Image acquisition and contour boundary extraction are performed on the completed Braille dot matrix, and the Braille dot diameter measurement dataset is calculated; The Braille dot height measurement dataset and the Braille dot diameter measurement dataset are respectively compared with the standard height value and the standard diameter value to obtain a geometric deviation parameter set, and Braille dot forming quality deviation data are generated based on the geometric deviation parameter set.

9. The printing control method according to claim 1, characterized in that, The step of performing gradient analysis and compensation on the first actuator control signal based on the Braille dot forming quality deviation data to obtain the second actuator control signal for the next printing process includes: The time derivative of the Braille dot forming quality deviation data is calculated to obtain deformation gradient data, and a sliding window analysis is performed based on the deformation gradient data to obtain the result of the actuator performance degradation trend judgment. When the result of the actuator performance degradation trend judgment exceeds the preset degradation threshold range, the fault diagnosis process is initiated and an actuator performance compensation strategy selection instruction is generated. According to the actuator performance compensation strategy, the control gain parameter in the first actuator control signal is adaptively adjusted by the selection instruction to obtain the target control parameter set. The target control parameter set is combined with the actuator remaining life prediction data to generate the second actuator control signal for the next printing process.

10. A printing control system, characterized in that, For performing the printing control method as described in any one of claims 1-9, the printing control system comprises: The preprocessing module is used to preprocess the text image of the sign to obtain the character sequence of the text to be converted, and to generate an actuator drive instruction group based on the character sequence of the text to be converted. The status feedback module is used to synchronously drive the printing actuator to perform Braille dot imprinting operation based on the actuator drive instruction group, and collect imprinting force and position deviation signals in real time to obtain actuator operation status feedback data. The parameter adjustment module is used to input the actuator operating status feedback data into the dual-channel attention control network for dynamic parameter adjustment, so as to obtain the first actuator control signal and the completed Braille dot matrix; The measurement module is used to perform optical detection and geometric dimension measurement on the completed Braille dot matrix to obtain Braille dot forming quality deviation data; The compensation module is used to perform gradient analysis and compensation on the first actuator control signal based on the Braille dot forming quality deviation data, so as to obtain the second actuator control signal for the next printing process.

Images