Intelligent laser engraving detection system for electric engraving machine

Abstract

The invention discloses an intelligent laser engraving detection system for an electric engraving machine, which relates to the technical field of intelligent detection, realizes real-time monitoring, fault prediction and parameter optimization of the running state of the electric engraving machine, acquires multi-source data such as running, environment, makeup software and proofing effect through a data acquisition module, analyzes a long-term dependency relationship by adopting a Transformer model, and realizes the real-time monitoring of the running state, fault prediction and parameter optimization of the electric engraving machine. The optimization control module adjusts the carving speed and the vibration frequency according to a prediction result, the service life of a carving needle is prolonged, and then the white scraping phenomenon during printing is reduced, the image recognition module evaluates the quality of carving net points through a CNN, the quality is fed back to the optimization control module to form closed-loop control, in addition, the system has the self-adaptive optimization function, and the control precision is improved. Autonomous learning is achieved through the knowledge base and the evaluation screening module, the performance is continuously improved, the operation efficiency, the engraving quality and the equipment service life of the electric engraving machine are remarkably improved, and meanwhile the maintenance cost and resource consumption are reduced.

Description

Technical Field

[0001] The invention relates to the technical field of intelligent detection, in particular to an intelligent laser engraving detection system for an electric engraving machine. Background Art

[0002] During the engraving process of the electric engraving machine, due to problems such as needle loss and unreasonable engraving parameter settings, small dots appear at the blade entrance of the image and text, and the edge is jagged, resulting in a reduction in ink content during printing, causing white scraping and affecting production efficiency. Traditional detection methods mostly rely on manual experience, which is inefficient and inaccurate. In recent years, the application of artificial intelligence technology in the field of industrial inspection has gradually increased. By introducing image processing, deep learning and other technologies, partial automated detection of the engraving process has been achieved.

[0003] Chinese invention patent CN110175603B discloses a method, system and storage medium for engraved text recognition, the method comprising: collecting images of the object to be measured by a binocular camera, generating a depth map by convolution and deconvolution operations, extracting text areas by a region recommendation network, and performing text recognition based on a bidirectional long short-term memory network to obtain recognition results of engraved text. The technology combines binocular stereo vision with deep learning technology to improve the significance and efficiency of text recognition, without the need for manual detection, reducing costs and improving reliability, and can be widely used in the field of text recognition. The method mainly focuses on the recognition of engraved text, and does not involve monitoring of the operating status of the electric engraving machine or fault prediction, and cannot solve actual fault problems such as needle loss and whitening of the blade. In addition, its application is limited to text recognition, and lacks the function of real-time monitoring and optimization of engraving quality.

[0004] Chinese invention patent application CN116226758A discloses an engraving machine anomaly detection method based on improved particle filtering and convolutional networks, which is divided into two stages: offline training and online anomaly detection. In the offline training stage, an accurate mathematical model is established by collecting operating data, introducing an improved particle filtering algorithm and an expectation maximization algorithm, and a one-dimensional residual convolutional neural network is trained to achieve data classification; in the online detection stage, the noise sequence is estimated through the model and the data anomaly is judged. This method can mine the data correlation in the time dimension and improve the accuracy of anomaly detection. This technology only focuses on the anomaly detection of the engraving machine operation data, fails to predict and prevent specific fault types (such as needle loss or whitening), and does not provide real-time parameter optimization functions to reduce the occurrence of faults. In addition, it relies on a mathematical model trained offline, which may be difficult to adapt to the dynamic changes in the operating state of the electric engraving machine and lacks the ability of adaptive learning and optimization.

[0005] The above designs improve the efficiency of text recognition through image processing and deep learning technology, or improve the accuracy of anomaly detection through mathematical modeling, but there are still certain limitations, such as: Lack of fault prediction and prevention: The existing technology fails to effectively predict specific faults such as needle loss and blade scraping, and cannot take preventive measures. Lack of real-time optimization function: The existing method does not realize real-time adjustment of engraving parameters, and cannot optimize equipment performance according to the operating status. Insufficient adaptive ability: Relying on static models or offline training, it is difficult to cope with dynamic changes in the operation of the electric engraving machine. Incomplete quality control: Lack of real-time monitoring and closed-loop feedback mechanism of engraving quality cannot ensure the effect of continuous optimization. Summary of the invention

[0006] The purpose of the present invention is to overcome the deficiencies of the prior art and to provide an intelligent laser engraving detection system for an electric engraving machine to solve the above-mentioned problems.

[0007] The object of the present invention is achieved through the following technical solutions: an intelligent laser engraving detection system for an electric engraving machine, comprising: Data acquisition module, used to collect multi-source data of the electric engraving machine in real time, including the operation data of the electric engraving machine, the imposition software data, the environment data and the proofing effect data; The data preprocessing module is used to clean, transform and compress the collected multi-source data to generate a format suitable for machine learning model processing; A data encoding module is used to encode the pre-processed multi-source data into branch data chains, data chains and data columns in chronological order, wherein a branch data chain is a time series of each data source, a data chain is a combination of multiple branch data chains in chronological order, and a data column is a parallel combination of data from multiple branch data chains at the same time point; The high-dimensional vector conversion module uses the Transformer model to convert the data chain into a high-dimensional vector, captures the long-term dependency of time steps through the self-attention mechanism, and encodes the data column into a high-dimensional vector through a fully connected layer; The impact correlation calculation module quantifies the impact correlation between different data sources by calculating the cosine similarity of high-dimensional vectors; The fault prediction module adds a prediction head based on the output layer of the Transformer model, including a regression layer for needle loss prediction and a classification layer or regression layer for blade edge scraping prediction, which outputs the loss degree and scraping probability or degree respectively; Optimize the control module and adjust the engraving speed and vibration frequency of the electric engraving machine in real time according to the fault prediction results to extend the life of the engraving needle and reduce the whitening phenomenon; The image recognition module uses a convolutional neural network (CNN) to analyze the engraved dot image, evaluate the dot quality, and feed the evaluation results back to the optimization control module to achieve closed-loop control; The optimization process recording module is used to record the input data, prediction results, optimization actions and feedback results of each optimization process; Evaluation and screening module, which evaluates the effect of each optimization through defined evaluation indicators or reinforcement learning reward functions, and screens out the optimization process with better optimization effect; The knowledge base module is used to store the selected key features and corresponding actions of the optimization process, and to match the current data with the patterns in the knowledge base in real time during the working process, and automatically call the optimization strategy with the highest matching degree.

[0008] The data acquisition module collects the electric engraving machine operation data including the engraving needle usage time, the number of engraving lines, the engraving time, the engraving needle vibration state and frequency, and the engraving needle material composition data.

[0009] The data preprocessing module compresses the TIFF image to ensure that the text is not distorted.

[0010] The Transformer model in the high-dimensional vector conversion module includes multiple encoding layers and self-attention heads to generate time-step vector representations of the data chain.

[0011] The needle loss prediction regression layer and the blade edge scraping prediction classification layer or regression layer in the fault prediction module are trained with historical data and updated regularly to improve prediction accuracy.

[0012] The optimization control module automatically reduces the engraving speed when it predicts that the needle wear exceeds the preset threshold, and adjusts the vibration frequency when the risk of scratching is predicted.

[0013] The image recognition module uses CNN to analyze the dot size and edge jaggedness to determine whether the ink content is sufficient during use, and uses the judgment result as feedback to input the optimization control module.

[0014] The optimization process recording module stores each optimization process in a relational database, and the record content includes branch data chains, data chains, data columns, prediction results, adjusted parameters and image recognition evaluation results.

[0015] The evaluation and screening module uses the improvement in network quality or the reduction in failure rate as evaluation indicators to screen out the optimization process with the best optimization effect.

[0016] The knowledge base module calculates the similarity between the current data and the storage optimization process, automatically calls the optimization strategy with the highest matching degree, and realizes adaptive learning and optimization.

[0017] The beneficial effects of the present invention are: 1. The system acquires multi-source data of the electric engraving machine in real time through the data acquisition module, including operation data, environmental data, imposition software data and proofing effect data. The Transformer model is used for high-dimensional vector conversion and self-attention mechanism analysis, which can capture the long-term dependency between data, thereby significantly improving the accuracy of fault prediction. Compared with the traditional maintenance method that relies on manual experience, the present invention can identify problems such as needle loss and blade scraping in advance, and the prediction error rate is greatly reduced. The optimization control module adjusts the engraving speed and vibration frequency in real time according to the fault prediction results, effectively prolonging the service life of the engraving needle and reducing the scraping phenomenon. At the same time, the image recognition module uses a convolutional neural network (CNN) to analyze the engraving dot image, evaluate the dot quality and feedback to the optimization control module to form a closed-loop control to ensure the effective implementation of the optimization measures. The data encoding module encodes the multi-source data into branch data chains, data chains and data columns in chronological order, and realizes structured processing and feature extraction of data through high-dimensional vector conversion and influence correlation calculation, which not only improves the input quality of the machine learning model, but also improves the training efficiency.

[0018] 2. Through predictive maintenance, the present invention can reduce the occurrence of sudden failures and emergency repairs, significantly reduce the company's maintenance costs, and at the same time, extend the service life of the engraving needle, reduce the replacement frequency and consumables expenditure, and save operating expenses for the company. The system automatically adjusts the operating parameters and reduces manual intervention, so that the electric engraving machine can continue to operate in the best state, which not only improves production efficiency, but also improves equipment utilization and reduces resource waste.

[0019] 3. By optimizing operating parameters and extending equipment life, the present invention reduces resource consumption and waste generation, conforms to the trend of green manufacturing and sustainable development, and contributes to environmental protection. The system reduces unexpected downtime and emergency repairs caused by faults, improves the safety of operators, and reduces potential risks in the working environment.

[0020] 4. The present invention applies the Transformer model and CNN technology to the field of electric engraving machine detection, realizes intelligent analysis and adaptive optimization of multi-source data, and through the knowledge base module and the evaluation and screening module, the system can autonomously learn and optimize and continuously improve performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 For the system interaction of the present invention Figure 1 ; Figure 2 is a timing diagram of the present invention; Figure 3 For the system interaction of the present invention Figure 2 . DETAILED DESCRIPTION

[0022] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0023] It is to be noted that the concepts of directions “left”, “right”, “up”, “down”, “front”, “back”, “inside” and “outside” in the following schemes are all relative directions and will not be listed one by one here.

[0024] Embodiment 1: like Figures 1 to 3 As shown, the following is a detailed description of Example 1, covering specific implementation methods, working processes and beneficial effects. This example aims to demonstrate the basic functional implementation of an intelligent laser engraving detection system for an electric engraving machine, including the coordinated operation of modules such as data acquisition, preprocessing, fault prediction and optimization control to achieve the goals of fault prediction and parameter optimization.

[0025] This embodiment provides an intelligent laser engraving detection system for an electric engraving machine. The system is composed of multiple modules. The specific implementation of each module is as follows: The data acquisition module is responsible for real-time collection of multi-source data of the electric engraving machine to comprehensively monitor the equipment operation status and environmental conditions. The specific collection contents include: Electric engraving machine operation data: collect the engraving needle usage time (reflecting the cumulative working time of the engraving needle), the number of engraving lines (such as 160, 180, 200, 230, 270, 330, 410, 540 and other different specifications), engraving time (the duration of a single task), the engraving needle vibration state and frequency (monitoring the dynamic characteristics of the engraving needle through a vibration sensor) and the engraving needle material composition data (recording the material characteristics of the engraving needle through material analysis) through sensors and equipment interfaces.

[0026] Imposition software data: Extract imposition parameters and layout information from imposition software (such as ARTCOM, Engraving make (CEM), Collage, HelioDisk) and analyze their impact on engraving quality.

[0027] Environmental data: The temperature and humidity in the workshop are collected through temperature and humidity sensors to evaluate the potential impact of environmental factors on equipment operation.

[0028] Proofing effect data: Obtain the official proofing results through the proofing machine, such as hue deviation or dot quality data, for subsequent quality evaluation.

[0029] The data preprocessing module processes the collected multi-source data and generates a format suitable for the machine learning model. The specific steps include: Data cleaning: fill in missing values ​​(for example, use interpolation to fill in missing points in the vibration frequency of the needle), remove or smooth outliers (for example, remove sudden peaks in the usage time of the needle).

[0030] Data conversion: Convert the raw data into a standardized format, such as one-hot encoding of categorical data (such as the type of needle material) and normalizing continuous data (such as vibration frequency).

[0031] Image compression: compress the TIFF images generated by the imposition software, using a lossless compression algorithm (such as LZW) to ensure that the text and details in the image are not distorted, while reducing the resources occupied by storage and transmission.

[0032] The fault prediction module implements fault prediction based on the Transformer model. The specific implementation is as follows: Model architecture: The Transformer model is used to process preprocessed multi-source data, and the self-attention mechanism is used to capture the long-term dependencies between data to generate high-dimensional vector representations.

[0033] Prediction head design: Add two prediction heads to the output layer of the Transformer model: Probe loss prediction: The degree of probe loss is output through the regression layer, expressed as a continuous value from 0 to 1 (0 means no loss, 1 means complete loss).

[0034] Prediction of blade whitening: Output the probability or degree of whitening through the classification layer or regression layer. For example, the severity of whitening is divided into three levels: mild, moderate, and severe, or the whitening risk value is output from 0 to 1.

[0035] Model training and updating: Use historical data (including multi-source data and corresponding fault records) to train the model, and regularly (for example, monthly) use newly collected data to update the model parameters to adapt to changes in equipment operating status and improve prediction accuracy.

[0036] The optimization control module adjusts the operating parameters of the electric engraving machine in real time according to the fault prediction results. The specific strategies include: Optimization of engraving needle loss: When the predicted engraving needle loss value exceeds the preset threshold (for example, 0.8), the system automatically reduces the engraving speed, for example, from 100% to 80%, to slow down the wear of the engraving needle.

[0037] Optimization of whitening risk: When the predicted probability of whitening exceeds a preset threshold (e.g. 0.5), the system adjusts the vibration frequency of the engraving needle, for example, from 500Hz to 450Hz, to reduce the occurrence of whitening at the blade entrance.

[0038] Parameter execution: The adjusted parameters are applied in real time through the control interface of the electric engraving machine to ensure that the optimization measures take effect immediately.

[0039] Working process The operation process of this system is a closed-loop real-time monitoring and optimization process. The specific steps are as follows: Data collection: During the operation of the electric engraving machine, the data acquisition module collects real-time operation data (engraving needle usage time, engraving line number, etc.), platemaking software data, environmental data and proofing effect data through sensors and software interfaces to form a multi-source data stream.

[0040] Data preprocessing: The collected multi-source data are sent to the data preprocessing module for cleaning (filling missing values, removing outliers), conversion (encoding and normalization) and compression (lossless compression of TIFF images) to generate clean input data with uniform format.

[0041] Data cleaning (interpolation): To fill missing values, use the linear interpolation formula: Data normalization (Min-Max normalization): Normalize the data using the following formula: Image compression (lossless compression): TIFF images generated by imposition software are compressed using the LZW algorithm, which uses a dictionary to generate lossless compressed files.

[0042] Failure prediction: The preprocessed data is input into the fault prediction module. The Transformer model analyzes the data and generates a high-dimensional vector. The prediction head outputs the degree of needle wear and the probability or degree of whitening as the fault prediction result.

[0043] Self-attention mechanism of Transformer model: The self-attention mechanism is used to capture the long-term dependencies between data. The calculation formula is as follows: Optimized control: The optimization control module receives the fault prediction results and determines whether the adjustment conditions are triggered: If the stylus wear exceeds a preset threshold, the engraving speed is reduced.

[0044] If the risk of whitening exceeds a preset threshold, the vibration frequency is adjusted.

[0045] The adjusted parameters are fed back to the electric engraving machine through the control system to achieve real-time optimization.

[0046] Loop execution: The system continuously collects new data and repeats the above process to form a closed-loop control of dynamic monitoring and optimization.

[0047] Probe loss prediction (regression layer output): The regression layer is used to predict the probe loss, and the output formula is: Prediction of whitening at the entry blade (output of the classification layer): The classification layer is used to determine the risk of whitening, using the softmax function: Engraving speed adjustment: When the needle loss prediction exceeds the threshold, the engraving speed is reduced. The following formula can be used: Vibration frequency adjustment: When the risk prediction of scraping exceeds the threshold, the vibration frequency is adjusted to: Closed-loop control and dynamic optimization: The system continues to run, and the control parameters are continuously optimized through the new data and fault prediction results collected in real time. For each prediction update, the objective function of the optimization control module can be expressed as: Dynamic parameter adjustment and optimization feedback: In the actual implementation process, through the feedback mechanism, the adjusted parameters will affect the operation of the equipment. This process can be achieved through the feedback control system formula: By integrating multi-source data through the Transformer model and making accurate predictions, the system can identify the risk of needle loss and whitening in advance, reduce unexpected downtime caused by faults, and improve equipment reliability.

[0048] Dynamically reduce the engraving speed based on the loss prediction, slow down the stylus wear rate, extend its service life, and thus reduce the replacement frequency and related costs.

[0049] By adjusting the vibration frequency, the incidence of white scraping at the blade entrance can be effectively reduced, the quality of engraved dots can be ensured, and the visual effect and market competitiveness of printed products can be improved.

[0050] The system's automated parameter adjustment reduces the need for manual intervention, allowing the electric engraving machine to continue to operate in the best condition, significantly improving production efficiency.

[0051] Predictive maintenance can reduce the occurrence of sudden failures and emergency repairs, reduce the company's maintenance costs and resource waste, and provide guarantees for long-term stable operation.

[0052] This embodiment realizes the intelligent fault prediction and operation optimization of the electric engraving machine through the coordinated work of the data acquisition module, the data preprocessing module, the fault prediction module and the optimization control module. In the specific implementation, the system pays attention to detail processing (such as lossless compression of TIFF images) and dynamic adjustment (such as real-time optimization of speed and frequency).

[0053] Embodiment 2: like Figures 1 to 3 As shown, based on Example 1, this embodiment further realizes the detailed functions of the data encoding module, the high-dimensional vector conversion module, the influence correlation calculation module, the fault prediction module and the related optimization module. Through these modules, the system can efficiently encode the multi-source data of the electric engraving machine into a high-dimensional vector, quantify the influence correlation between the data sources, and combine image recognition and knowledge base optimization to achieve accurate fault prediction and intelligent parameter adjustment, thereby improving the operating efficiency and engraving quality of the electric engraving machine.

[0054] The data acquisition module collects multi-source data of the electric engraving machine in real time, including: Electric engraving machine operation data: such as needle usage time, engraving speed, vibration frequency, etc.; Imposition software data: such as engraving line number, dot design parameters, etc. Environmental data: such as temperature, humidity, etc.; Proofing effect data: such as preliminary evaluation results of dot quality.

[0055] The preprocessing module cleans (such as removing noise), transforms (such as normalization) and compresses (such as dimensionality reduction or feature extraction) the collected multi-source data to generate a format suitable for machine learning model processing. For example, the needle usage time and ambient temperature are normalized to the range of 0 to 1 to facilitate subsequent encoding and modeling.

[0056] The data encoding module encodes the preprocessed multi-source data in chronological order into the following three forms: Branch data chain: Each data source forms a time series. For example, the branch data chain of the usage time of the engraving needle is expressed as [T1, T2, T3, ...], where T1 is the usage time of the engraving needle at a certain moment.

[0057] Data chain: Multiple branch data chains are combined into a complete sequence in chronological order, such as [T1, L1, E1, T2, L2, E2, ...], where L represents the number of engraving lines and E represents the ambient temperature.

[0058] Data column: Multi-source data at the same time point is combined in parallel, such as the data column at time t1 is (T1, L1, E1).

[0059] This encoding method retains the temporal characteristics of the data (data chain) and multi-source correlation (data column), providing structured input for high-dimensional vector conversion.

[0060] The high-dimensional vector conversion module uses the Transformer model to process data chains and data columns to generate high-dimensional vector representations: Model architecture: The Transformer model contains multiple encoding layers and self-attention heads. The self-attention mechanism captures long-term dependencies in the data chain by calculating the weights between time steps. For example, it analyzes the correlation between the length of time the engraving needle is used and the number of engraving lines at different time points.

[0061] Data chain processing: Input the data chain into the Transformer model to generate a high-dimensional vector representation for each time step, which contains temporal context information.

[0062] Data column processing: Encode the data column into a high-dimensional vector through a fully connected layer to capture the joint features of multi-source data at the same time.

[0063] The final high-dimensional vector output combines time series and multi-source characteristics, providing a rich information basis for subsequent analysis.

[0064] The impact correlation calculation module quantifies the impact correlation between different data sources by calculating the cosine similarity of high-dimensional vectors: Calculation method: For two data sources (such as the length of time the engraving needle is used and the ambient temperature), extract their high-dimensional vectors and calculate the cosine similarity. The result is between -1 and 1. The closer the value is to 1, the higher the correlation.

[0065] Application example: If the similarity between the needle usage time and the ambient temperature is high, it means that temperature has a significant impact on needle loss, which can be used to optimize the weight of the prediction model.

[0066] The fault prediction module adds a prediction head to the output layer of the Transformer model to implement two types of fault prediction: Probe loss prediction: Output the loss degree through the regression layer (0 to 1 continuous value, 0 means no loss, 1 means complete loss).

[0067] Prediction of blade whitening: The classification layer outputs the severity of whitening (such as none, mild, moderate, severe) or the regression layer outputs the risk value of whitening (0 to 1).

[0068] Training and updating: Use historical data to train the model and update parameters regularly (e.g. monthly) to adapt to changes in equipment status.

[0069] The optimization control module adjusts the parameters of the electric engraving machine in real time according to the fault prediction results: If the needle loss is predicted to be high, reduce the engraving speed to extend the needle life; If the risk of white scraping is predicted to be high, adjust the vibration frequency to reduce white scraping.

[0070] The image recognition module uses a convolutional neural network (CNN) to analyze the engraved dot image, evaluate the dot quality (such as uniformity and edge clarity), and feed the results back to the optimization control module to form a closed-loop control. For example, if the dot edge is blurred, it prompts to adjust the vibration frequency.

[0071] The optimization process recording module records the input data (such as high-dimensional vectors), prediction results (such as loss degree), optimization actions (such as speed reduction) and feedback results (such as improvement in network point quality) of each optimization.

[0072] The evaluation and screening module evaluates the optimization effect through evaluation indicators (such as prediction accuracy, improvement in network quality) or reinforcement learning reward functions, and screens out the optimization process with the best effect.

[0073] The knowledge base module stores the selected key features of the optimization process (such as data patterns under high-loss conditions) and corresponding actions (such as reducing speed), and matches the current data in real time while working to call the optimal strategy.

[0074] Working process Data collection and preprocessing The system collects data such as the usage time of the engraving needle, the number of engraving lines, and the ambient temperature in real time, and generates structured data after cleaning and normalization.

[0075] Data Encoding Data is encoded as data links (e.g. [T1, T2, T3, ...]), data links (e.g. [T1, L1, E1,T2, L2, E2, ...]) and data columns (e.g. (T1, L1, E1)).

[0076] Branch data chain (time series encoding): Branch data chain is represented as a time series, and the following dynamic encoding function can be used: Data chain (multi-source data sequence encoding): The time series of multiple data sources are jointly encoded, and the following formula can be defined: Data column (parallel combination): Multi-source data at the same time is combined in parallel into a data column, which is defined as: High Dimensional Vector Transformation The data chain is processed by the Transformer model, and the self-attention mechanism captures temporal dependencies and generates high-dimensional vectors.

[0077] The data columns are encoded through a fully connected layer to generate multi-source feature vectors.

[0078] Data chain processing (Transformer model application): Apply the Transformer model to the data chain to generate a high-dimensional vector representation for each time step. Assuming that each data source passes through the self-attention mechanism, a high-dimensional vector H is obtained for each time step. i , can be generated by the following formula: Data column processing (fully connected layer encoding): Encode the data column and convert the multi-source features at each moment into a high-dimensional vector, defined as: Impact correlation calculation Calculate the cosine similarity of high-dimensional vectors. For example, the similarity between the length of time the engraving needle is used and the number of engraving lines is 0.85, indicating that the two are highly correlated.

[0079] Cosine similarity calculation: It is used to quantify the correlation between different data sources. For two data sources (such as the needle usage time T and the ambient temperature E), their cosine similarity can be calculated by the following formula: Assuming that the similarity between the needle usage time and the ambient temperature is high (for example, 0.85), it means that temperature has a significant impact on needle loss, and the model should increase the weight of temperature data.

[0080] Failure prediction The Transformer output layer predicts the needle loss (such as 0.75, close to the threshold) and the risk of scratching (such as 0.6, medium risk).

[0081] Probe loss prediction (regression layer output): Use the regression layer to predict the probe loss, the output formula is: Prediction of whitening at the inlet edge (classification layer output): Classify or regress the risk of whitening and use the softmax function to output the probability: Parameter Optimization The optimization control module reduces the engraving speed and adjusts the vibration frequency according to the prediction results, while the image recognition module analyzes the dot quality and provides feedback on the adjustment effect.

[0082] Parameter optimization (speed and frequency adjustment): Adjust the parameters according to the fault prediction results. If the loss prediction is high, reduce the engraving speed. Assuming the current speed is v current , the optimized speed is: Dot quality assessment: Use convolutional neural network (CNN) to analyze the engraved dot image and generate the assessment result. Assume that I dot is a dot image, and the evaluation function can be expressed as: Recording and Optimization The optimization process recording module stores the operation data, the evaluation and screening module evaluates the effect, and the knowledge base module updates the optimization strategy.

[0083] Optimization process record: Record the input data, prediction results, optimization actions and feedback results of each optimization. The optimization process can be achieved through the following recording functions: Knowledge base update and call: Store the key features and corresponding actions of the optimization process, and call the optimal strategy in real time. The update process can be represented by the following function: Continuous operation The system cycles the above steps to achieve dynamic monitoring and optimization.

[0084] The data encoding module and high-dimensional vector conversion module comprehensively capture the timing and multi-source characteristics through data branches, data chains and Transformer models, thereby improving data utilization efficiency.

[0085] Cosine similarity calculation quantifies the correlation between data sources and provides a scientific basis for fault prediction and parameter adjustment.

[0086] The Transformer model combined with the prediction head and periodic update mechanism significantly improves the accuracy of needle loss and scratch white prediction.

[0087] The optimization control module and the image recognition module form a closed-loop control, adjust the parameters in real time, extend the life of the engraving needle and improve the engraving quality.

[0088] The knowledge base module automatically calls the optimal strategy through historical data learning and real-time matching, thereby improving the intelligence level of the system.

[0089] Embodiment 2 realizes efficient processing and intelligent optimization of multi-source data of electric engraving machines through the collaborative work of modules such as data encoding, high-dimensional vector conversion, impact correlation calculation and fault prediction.

[0090] Embodiment 3: like Figures 1 to 3 As shown, this embodiment further realizes the functions of image recognition module, optimization process recording module, evaluation and screening module and knowledge base module on the basis of embodiment 1 (data collection, preprocessing, fault prediction and optimization control) and embodiment 2 (data encoding, high-dimensional vector conversion and influence correlation calculation). Through these modules, the system can evaluate the quality of engraving dots in real time, record the optimization process, screen high-quality optimization strategies, and realize adaptive learning through the knowledge base, so as to optimize the operating parameters of the electric engraving machine and improve the level of intelligence.

[0091] The image recognition module is based on convolutional neural network (CNN) technology, which is used to analyze the image of engraved dots, evaluate the dot quality, and feed the results back to the optimization control module. The implementation details are as follows: Dot image acquisition: After the electric engraving machine completes engraving, the dot image is captured in real time by a high-resolution camera and transmitted to the image recognition module.

[0092] Image preprocessing: grayscale the captured image to reduce noise interference, reduce noise through Gaussian filtering, and use edge enhancement technology to highlight the dot outline and details.

[0093] CNN model architecture: A multi-layer CNN model is constructed, including convolutional layers (to extract dot features, such as size and edge shape), pooling layers (to reduce computational complexity), and fully connected layers (to output quality assessment results). The model learns dot features through training, such as uniformity of dot size, edge jaggedness, and shape consistency.

[0094] Network point quality assessment: The model outputs quantitative indicators, including: Dot size uniformity (measured by calculating the standard deviation, the smaller the value, the better); The degree of edge jaggedness (the less jagged the better); Whether the ink content is sufficient (judged by the fullness of the dots, for example, insufficient fullness may indicate insufficient ink content during printing).

[0095] Feedback mechanism: Feedback the evaluation results (such as dot quality score or insufficient ink content) to the optimization control module. For example, if the edge is too jagged, it may trigger the adjustment of vibration frequency; if the ink content is insufficient, it may prompt to adjust the engraving speed.

[0096] The optimization process recording module is responsible for recording the detailed data of each optimization to provide support for subsequent analysis and learning. The specific implementation is as follows: Record content: including: Support data chain: such as time series data of the length of time the engraving needle is used; Data chain: Multi-source data (such as operation data and environmental data) are combined in chronological order; Data column: parallel combination of multi-source data at the same time; Prediction results: such as the needle loss value and scratch risk output by the fault prediction module; Adjust parameters: such as specific changes in engraving speed and vibration frequency; Image recognition evaluation results: such as dot quality score.

[0097] Storage method: Use a relational database (such as MySQL) to store optimization records, and design an "optimization record table" containing fields such as "timestamp", "branch data chain", "prediction results", "adjustment parameters", "outlet quality score", etc. to facilitate query and analysis.

[0098] The evaluation and screening module analyzes the optimization effect through evaluation indicators and screens out high-quality optimization processes. The specific implementation is as follows: Evaluation Metrics: Improvement of outlet quality: Calculate the difference between the outlet quality scores before and after the adjustment; Fault reduction rate: such as the percentage reduction of white scraping phenomenon after adjustment.

[0099] Reinforcement learning mechanism (optional): Using reinforcement learning algorithm, optimization actions are regarded as actions, and the improvement of network quality is regarded as reward, and the optimal strategy is learned through trial and error.

[0100] Screening criteria: Set a threshold (such as the improvement in dot quality exceeds 10% or the reduction in white spot scraping exceeds 15%), screen the optimization processes that meet the conditions, and mark them as "high-quality optimization".

[0101] The knowledge base module stores high-quality optimization processes and supports adaptive optimization. The specific implementation is as follows: Knowledge base construction: The key features of the selected high-quality optimization process (such as data patterns under high-loss conditions) and corresponding actions (such as reducing the engraving speed by 10%) are stored in the knowledge base.

[0102] Similarity calculation: During the operation, the Euclidean distance or cosine similarity is used to calculate the similarity between the current data and the optimization process in the knowledge base.

[0103] Automatic call: When the similarity exceeds the preset threshold (such as 0.9), the optimization strategy with the highest matching degree is automatically called and applied to the current operation to achieve adaptive optimization.

[0104] Working process The data acquisition module acquires the operation data and environmental data of the electric engraving machine in real time. After being cleaned, converted and compressed by the data preprocessing module, the data encoding module generates branch data chains, data chains and data columns, and then generates input features through the high-dimensional vector conversion module.

[0105] Fault prediction and preliminary optimization The fault prediction module outputs the degree of needle wear and the risk of whitening based on the Transformer model. The optimization control module adjusts the engraving speed or vibration frequency according to the prediction results to extend the life of the needle and reduce whitening.

[0106] Image Recognition Evaluation After engraving is completed, the image recognition module analyzes the dot image and outputs the dot quality score and ink content assessment. If the quality does not meet the standard (such as too much jagged edges), the results are fed back to the optimization control module to trigger further adjustments.

[0107] Optimize records The optimization process recording module stores the input data, prediction results, adjustment parameters and network point quality scores of this optimization in a relational database.

[0108] Evaluation and screening The evaluation and screening module calculates the optimization effect (such as the improvement in the quality of the network points). If it exceeds the threshold (such as 10%), it is marked as a high-quality optimization.

[0109] Knowledge base update and call The high-quality optimization process is stored in the knowledge base module. In subsequent operations, the system calculates the similarity between the current data and the knowledge base model in real time and automatically calls the optimization strategy with the highest matching degree.

[0110] Continuous Optimization The system forms an adaptive optimization closed loop through recording, evaluation and learning, and continuously improves operating results.

[0111] The image recognition module uses CNN to analyze dot size, edge jaggedness and ink content, objectively quantifying the engraving quality and improving the accuracy and reliability of quality control.

[0112] The image recognition results are fed back to the optimization control module in real time, forming a closed loop of parameter adjustment and quality assessment to ensure the effective implementation of optimization measures.

[0113] The optimization process recording module stores detailed data of each optimization in a relational database, providing a rich data foundation for subsequent improvements and problem tracing.

[0114] The knowledge base module achieves system adaptive learning through similarity matching and automatic invocation of optimization strategies, reduces manual intervention, and improves the level of intelligence.

[0115] The evaluation and screening module selects high-quality optimization processes and stores them in the knowledge base, ensuring that the system continuously accumulates experience and continuously improves the operating efficiency and engraving quality of the electric engraving machine.

[0116] Example 3 realizes real-time evaluation of engraving quality and adaptive optimization of operating parameters through the collaborative work of the image recognition module, the optimization process recording module, the evaluation and screening module and the knowledge base module. On the basis of Examples 1 and 2, this example adds closed-loop control and intelligent learning capabilities, thereby improving the intelligence and practicality of the electric engraving machine detection system.

[0117] The above is only a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein and should not be regarded as excluding other embodiments. It can be used in various other combinations, modifications and environments, and can be modified within the scope of the concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art do not depart from the spirit and scope of the present invention, and should all be within the scope of protection of the claims attached to the present invention.

Claims

1. An intelligent laser engraving detection system for an electric engraving machine, characterized in that: include: Data acquisition module, used to collect multi-source data of the electric engraving machine in real time; The data preprocessing module is used to clean, transform and compress the collected multi-source data to generate a format suitable for machine learning model processing; A data encoding module is used to encode the pre-processed multi-source data into branch data chains, data chains and data columns in chronological order; The high-dimensional vector conversion module uses the Transformer model to convert the data chain into a high-dimensional vector, captures the long-term dependency of time steps through the self-attention mechanism, and encodes the data column into a high-dimensional vector through a fully connected layer; The impact correlation calculation module quantifies the impact correlation between different data sources by calculating the cosine similarity of high-dimensional vectors; The fault prediction module adds a prediction head based on the output layer of the Transformer model; Optimize the control module and adjust the engraving speed and vibration frequency of the electric engraving machine in real time according to the fault prediction results to extend the life of the engraving needle and reduce the whitening phenomenon; The image recognition module uses a convolutional neural network (CNN) to analyze the engraved dot image, evaluate the dot quality, and feed the evaluation results back to the optimization control module to achieve closed-loop control.

2. The intelligent laser engraving detection system for an electric engraving machine according to claim 1, characterized in that: It also includes an optimization process recording module, an evaluation and screening module and a knowledge base module. The evaluation and screening module evaluates each optimization effect through a defined evaluation index or a reinforcement learning reward function, and screens out an optimization process with a better optimization effect; The knowledge base module is used to store the selected key features and corresponding actions of the optimization process, and to match the current data with the patterns in the knowledge base in real time during the working process, and automatically call the optimization strategy with the highest matching degree; the multi-source data includes the operation data of the electric engraving machine, the imposition software data, the environmental data and the proofing effect data, and the operation data of the electric engraving machine collected by the data acquisition module includes the use time of the engraving needle, the number of engraving lines, the engraving time, the vibration state and frequency of the engraving needle and the composition data of the engraving needle material.

3. The intelligent laser engraving detection system for an electric engraving machine according to claim 1 is characterized in that: The branch data chain is a time series of each data source, the data chain is a combination of multiple branch data chains in time sequence, the data column is a parallel combination of data from multiple branch data chains at the same time point, and the data preprocessing module compresses the TIFF image to ensure that the text is not distorted.

4. The intelligent laser engraving detection system for an electric engraving machine according to claim 1 is characterized in that: The fault prediction module includes a regression layer for predicting needle loss and a classification layer or regression layer for predicting whitening of the blade, which output the degree of loss and the probability or degree of whitening respectively. The Transformer model in the high-dimensional vector conversion module includes multiple encoding layers and self-attention heads, which are used to generate a time step vector representation of the data chain.

5. The intelligent laser engraving detection system for an electric engraving machine according to claim 1 is characterized in that: The optimization process recording module is used to record the input data, prediction results, optimization actions and feedback results of each optimization process. The needle loss prediction regression layer and the blade edge scraping prediction classification layer or regression layer in the fault prediction module are trained with historical data and updated regularly to improve the prediction accuracy.

6. The intelligent laser engraving detection system for an electric engraving machine according to claim 5 is characterized in that: The optimization control module automatically reduces the engraving speed when predicting that the engraving needle loss exceeds a preset threshold, and adjusts the vibration frequency when predicting the risk of whitening.

7. The intelligent laser engraving detection system for an electric engraving machine according to claim 1, characterized in that: The image recognition module analyzes the dot size and edge jaggedness through CNN to determine whether the ink content is sufficient during use, and inputs the judgment result as feedback to the optimization control module.

8. The intelligent laser engraving detection system for an electric engraving machine according to claim 1 is characterized in that: The optimization process recording module stores each optimization process in a relational database, and the recorded content includes branch data chains, data chains, data columns, prediction results, adjusted parameters and image recognition evaluation results.

9. The intelligent laser engraving detection system for an electric engraving machine according to claim 1, characterized in that: The evaluation and screening module uses the improvement of network point quality or the failure reduction rate as an evaluation index to screen out the optimization process with the best optimization effect.

10. The intelligent laser engraving detection system for an electric engraving machine according to claim 2, characterized in that: The knowledge base module automatically calls the optimization strategy with the highest matching degree by calculating the similarity between the current data and the storage optimization process, thereby achieving adaptive learning and optimization.

Images