Bus type pipe cutting system real-time control method based on digital twinning

Abstract

The invention discloses a bus type pipe cutting system real-time control method based on digital twinning. The bus type pipe cutting system real-time control method comprises the following steps of 1, collecting and preprocessing multi-dimensional sensor data; 2, performing high-order differential processing based on the standardized sensor data set; 3, performing autoregression modeling according to the differential time sequence data to obtain a global time sequence prediction result; 4, collecting pipe cutting image data, and inputting the pipe cutting image data into the ResNeXt network for packet convolution and residual connection; 5, fusing and inputting the global time sequence prediction result and the cut tube image feature vector into an improved DeepAR model for variational reasoning and Bayesian estimation; 6, adjusting the control parameters according to the posterior distribution of the prediction result, and generating a corresponding control instruction; and 7, monitoring the difference value between the actual operation data and the prediction result, and correcting the control parameters. According to the invention, the ResNeXt network and the improved DeepAR model are fused, and the real-time accurate control of the cutting and management system is realized.

Description

Technical Field

[0001] This invention relates to the field of real-time control technology, and in particular to a real-time control method for a bus-type pipe cutting system based on digital twins. Background Technology

[0002] As the manufacturing industry moves towards intelligence, digitalization, and automation, the demands for precision control in industrial production are constantly increasing. This is especially true in pipe cutting systems, where traditional experience-based control methods are increasingly inadequate to meet the demands of complex cutting processes. Traditional pipe cutting systems typically rely on manual intervention or fixed control rules to adjust cutting parameters such as cutting speed, feed rate, cutting path, and tool pressure. These methods often have significant limitations. With changes in the production environment, traditional methods struggle to adjust cutting parameters in real time, and they often fail to maintain consistent cutting quality when handling pipes of different materials and thicknesses. Therefore, existing technologies face certain technical bottlenecks in improving cutting efficiency and accuracy, particularly when dealing with complex cutting processes and uncertainties, where traditional methods exhibit poor adaptability and robustness.

[0003] In existing technologies, many pipe cutting systems still rely on simple sensor inputs and preset control rules. While these methods can guarantee basic cutting operations, they cannot adjust in real time to address complex and dynamically changing factors during the cutting process (such as tool wear, material reflectivity, and changes in the cutting edge). This often results in suboptimal control of the cutting process in actual production, leading to fluctuations in cutting accuracy and even scrap and equipment damage. Furthermore, although some pipe cutting systems have introduced model-based control strategies in recent years, these methods typically rely on static models and cannot effectively cope with real-time changing conditions. For example, some methods use fixed machine learning models to predict cutting parameters, but because the models fail to update dynamically based on real-time data, there is a significant error between the predictions and the actual operation. This error not only affects cutting quality but also increases production costs and reduces equipment lifespan. Therefore, there is an urgent need for a more intelligent and dynamic control method that can comprehensively utilize real-time sensor data and image data, and automatically adjust control parameters based on feedback information during the cutting process, thereby ensuring the accuracy and stability of the cutting process. Furthermore, existing technologies often neglect how to quantify the uncertainty of prediction results, making it difficult for control decisions to achieve robustness when faced with complex and dynamic operating conditions.

[0004] Therefore, how to provide a real-time control method for a bus-type pipe cutting system based on digital twins is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a real-time control method for a bus-type pipe cutting system based on digital twins. This invention achieves real-time and precise control of the bus-type pipe cutting system by fusing multi-dimensional sensor data and image data, combined with an improved DeepAR model and ResNeXt network. High-order differential and piecewise autoregressive modeling methods are employed to dynamically adjust control parameters, ensuring high precision and stability in the cutting process. Simultaneously, variational inference and Bayesian estimation are used to quantify the uncertainty of prediction results, improving the system's robustness and reliability in uncertain environments, and significantly enhancing cutting quality and production efficiency.

[0006] A real-time control method for a bus-type pipe cutting system based on digital twins according to an embodiment of the present invention includes the following steps: Step 1: Collect multidimensional sensor data and preprocess the multidimensional sensor data to obtain a standardized sensor dataset; Step 2: Perform high-order differencing based on the standardized sensor dataset to obtain differential time series data; Step 3: Divide the differential time series data into segments according to a preset time period, perform independent autoregressive modeling in each time period, and fuse the autoregressive prediction results of each time period to obtain the global time series prediction result. Step 4: Real-time acquisition of pipe-cutting image data during the cutting process, and inputting the pipe-cutting image data into the ResNeXt network to obtain the pipe-cutting image feature vector through grouped convolution and residual connections; Step 5: Fuse the global time series prediction results and the tube-cutting image feature vector to generate a joint feature vector, and input the joint feature vector into the improved DeepAR model. Through variational inference and Bayesian estimation, the posterior distribution of the prediction results is obtained. Step 6: Based on the posterior distribution of the predicted results, calculate and adjust the control parameters in real time, and generate corresponding control commands; Step 7: Monitor the difference between the actual operating data and the predicted results during the cutting process. When the deviation exceeds the preset threshold, adjust the control parameters.

[0007] Optionally, step one specifically includes: The speed of the cutting tool and the feed rate as it moves across the pipe surface are monitored in real time using a speed sensor. Temperature sensors are used to acquire temperature data of the cutting area and the cutting tool. The cutting force or pressure during the cutting process is detected by a pressure sensor; Vibration sensors are used to collect the vibration status of the cutting tool and the wear status of the cutting tool. Remove outliers from the multidimensional sensor data and fill in the missing values ​​in the sensor data using an average interpolation method; A low-pass filter is used to smooth the multi-dimensional sensor data and remove high-frequency noise; a high-pass filter is used to remove low-frequency noise. Different types of data in the multidimensional sensor data are normalized separately to obtain a standardized sensor dataset.

[0008] Optionally, step two specifically includes: The standardized sensor dataset is sorted in chronological order to obtain time series data; Higher-order differencing is performed on the time series data to obtain differencing time series data. The specific steps of higher-order differencing are as follows: The original time series data is subjected to a first difference, which is to subtract the data of the previous time point from the data of each time point to obtain a new sequence. The new sequence is subjected to a second difference, which is to subtract the data of the previous time point from each time point in the new sequence to obtain the second difference data. Repeat the iteration until the difference between any two adjacent time points is less than a preset threshold to obtain differential time series data.

[0009] Optionally, step three specifically includes: The differential time series data is segmented according to a preset time period length, with each time period containing a fixed number of data points; A sliding time window is used to progressively divide each time period into multiple time period subsets; Independent autoregressive modeling is performed on each time period subset to obtain the autoregressive prediction results for each time period subset. The specific autoregressive modeling steps are as follows: The order of autoregressive modeling is determined for a subset of time periods using the Akaike information content criterion, where the order is determined by the number of past time points used for prediction. The least squares method is used to estimate the autoregressive coefficients in the autoregressive model and to minimize the error between the predicted and actual values. The autoregressive model is used to calculate the predicted values ​​for future time points to obtain the autoregressive prediction results. The autoregressive prediction results of the subsets of data from each time period are weighted and averaged to obtain the global time series prediction result.

[0010] Optionally, step four specifically involves: Real-time acquisition of tube cutting image data during the cutting process using high-definition cameras or infrared sensors; The tube cutting image data includes tool wear, cutting edge, and material reflectivity; the tool wear includes the wear depth and morphological change characteristics of the tool. The cut edge includes the flatness and serration of the edge; the material reflectance is the surface smoothness of the material; The acquired tube-cutting image data is input into a ResNeXt network for feature extraction. The ResNeXt network extracts key information from the image in the following way: The tube-cutting image data is input into a ResNeXt network, and the tube-cutting image feature vector is obtained through grouped convolution and residual connections. The specific steps are as follows: ResNeXt divides the tube-cutting image data into multiple groups using grouped convolution technology. Each group is convolved using an independent convolution kernel, and multiple residual blocks are stacked. Each residual block contains a skip connection, multiple convolutional layers, and a ReLU activation function. The skip connection allows the input data to skip the convolutional layer and be passed directly to the pooling layer; the ReLU activation function performs a non-linear transformation, setting all negative values ​​to zero and keeping positive values ​​unchanged. Local features in the tube-cutting image are extracted by convolution operation. These local features include the depth of tool wear, morphological changes, and the shape of the cutting edge. Global features of the tube-cutting image are extracted through multiple convolutional and pooling layers to capture the overall structure in the tube-cutting image, including changes in the cutting surface quality and cutting angle; the pooling layer downsamples local features through max pooling operation. The global features are sorted according to the time series to obtain the feature vector of the tube-cutting image.

[0011] Optionally, step five specifically includes: The global time series prediction results and the feature vector of the tube-cutting image are concatenated along the feature dimension. The splicing step involves aligning the two vectors element by element, aligning them element by element along the feature dimension, and then connecting the two vectors to generate a long vector containing time series and tube-cutting image information, which serves as the final joint feature vector. The joint feature vector is input into the improved DeepAR model, and the posterior distribution of the prediction results is obtained through variational inference and Bayesian estimation. The specific steps are as follows: The prior distribution and marginal likelihood are obtained from the joint eigenvectors using the Laplace distribution; The posterior distribution is derived from the prior distribution and marginal likelihood using Bayes' theorem. Variational posterior distribution is obtained by using variational inference on the posterior distribution, wherein the variational inference involves selecting a Gaussian distribution as a family of distributions, and the family of distributions is used to approximate the variational posterior distribution. Based on the variational posterior distribution, Bayesian estimation is used to update the value of each prediction parameter to obtain the posterior distribution of the prediction result. The Bayesian estimation steps are as follows: Calculate the posterior expectation of each parameter in the variational posterior distribution to obtain the predicted value of each parameter; Calculate the variance or standard deviation of the variational posterior distribution, provide the confidence interval for the predicted value of each parameter, and obtain the posterior distribution of the prediction result. The confidence interval represents the fluctuation range of the predicted value.

[0012] Optionally, step six specifically includes: Based on the posterior distribution of the prediction results, the most probable value and the corresponding confidence interval of each prediction control parameter are extracted. The prediction control parameters include cutting speed, feed speed, feed angle, cutting pressure and cutting path. The most probable value is the expected value of the corresponding posterior distribution. Adjust the tool feed rate according to the predicted feed rate to increase the material removal rate during the cutting process; By adjusting various predictive control parameters, corresponding control commands are generated, including motor control commands and tool control commands.

[0013] Optionally, step seven specifically includes: The actual operating data is obtained by real-time monitoring of various data during the cutting process using sensors, including cutting speed, feed rate, cutting pressure, tool wear, and temperature parameters. The actual operating data monitored are compared with the previously predicted control parameters, and the difference is calculated. If the difference exceeds a preset threshold, the control parameters are adjusted according to the magnitude of the deviation. The adjustment involves proportionally scaling the cutting speed, feed rate, cutting pressure, and tool angle.

[0014] The beneficial effects of this invention are: This invention significantly improves the control accuracy and stability of a bus-type pipe cutting system by combining digital twin technology with a deep learning model. By acquiring multi-dimensional sensor data and image data during the cutting process in real time, and using an improved DeepAR model and a ResNeXt network for data fusion and prediction, this method can adjust control parameters such as cutting speed, feed rate, and cutting path in real time, ensuring precise control of the cutting process. Traditional pipe cutting systems often employ static control rules or experience-based prediction methods, which cannot flexibly cope with different cutting conditions and dynamically changing requirements. This invention, through innovative technologies such as high-order difference and piecewise autoregressive modeling, enables the system to capture complex nonlinear characteristics and local variation patterns in the data, thereby dynamically adjusting various control parameters during the cutting process to adapt to different working conditions.

[0015] When factors such as tool wear, cutting edge quality, and material reflectivity vary significantly, the system of this invention can respond and adjust its control strategy promptly, avoiding cutting quality fluctuations caused by fixed rules or preset parameters in traditional methods. Through variational inference and Bayesian estimation, this invention can also quantify the uncertainty of prediction results and provide confidence intervals for each control parameter, thus providing more robust decision support for the system. This enables the system to make more reliable control decisions in production environments with high uncertainty, reducing losses due to prediction errors and improving production reliability and efficiency. This invention effectively enhances the monitoring capability of the cutting process by combining time-series data with image data through multimodal data fusion. By extracting image features through a ResNeXt network and fusing them with the prediction results of time-series data, the system can not only make predictions based on historical data but also acquire key information from images in real time, further improving its adaptability to various complex changes during the cutting process. This multi-dimensional fusion method gives this invention higher accuracy and flexibility than traditional methods when handling complex cutting processes. It significantly improves the intelligence level of the pipe cutting system, enhances the control precision, robustness and adaptability, and provides a more reliable and efficient technical solution for the manufacturing industry, especially the high-precision cutting industry. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0017] Figure 1 This is an overall flowchart of a real-time control method for a bus-type pipe cutting system based on digital twin proposed in this invention; Figure 2 This is a schematic diagram of the processing flow of the ResNeXt network in the real-time control method of a bus-type pipe cutting system based on digital twin proposed in this invention. Figure 3 This is a flowchart illustrating the processing steps of an improved DeepAR model for a real-time control method of a bus-type pipe cutting system based on digital twins, as proposed in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figure 1-3 A real-time control method for a bus-type pipe cutting system based on digital twins includes the following steps: Step 1: Collect multidimensional sensor data and preprocess the multidimensional sensor data to obtain a standardized sensor dataset; Step 2: Perform high-order differencing based on the standardized sensor dataset to obtain differential time series data; Step 3: Divide the differential time series data into segments according to a preset time period, perform independent autoregressive modeling in each time period, and fuse the autoregressive prediction results of each time period to obtain the global time series prediction result. Step 4: Real-time acquisition of pipe-cutting image data during the cutting process, and inputting the pipe-cutting image data into the ResNeXt network to obtain the pipe-cutting image feature vector through grouped convolution and residual connections; Step 5: Fuse the global time series prediction results and the tube-cutting image feature vector to generate a joint feature vector, and input the joint feature vector into the improved DeepAR model. Through variational inference and Bayesian estimation, the posterior distribution of the prediction results is obtained. Step 6: Based on the posterior distribution of the predicted results, calculate and adjust the control parameters in real time, and generate corresponding control commands; Step 7: Monitor the difference between the actual operating data and the predicted results during the cutting process. When the deviation exceeds the preset threshold, adjust the control parameters.

[0020] In this embodiment, step one specifically includes: The speed of the cutting tool and the feed rate as it moves across the pipe surface are monitored in real time using a speed sensor. Temperature sensors are used to acquire temperature data of the cutting area and the cutting tool. The cutting force or pressure during the cutting process is detected by a pressure sensor; Vibration sensors are used to collect the vibration status of the cutting tool and the wear status of the cutting tool. Remove outliers from the multidimensional sensor data and fill in the missing values ​​in the sensor data using an average interpolation method; A low-pass filter is used to smooth the multi-dimensional sensor data and remove high-frequency noise; a high-pass filter is used to remove low-frequency noise. Different types of data in the multidimensional sensor data are normalized separately to obtain a standardized sensor dataset.

[0021] This step achieves precise monitoring and optimized control of the pipe cutting system by comprehensively applying multi-dimensional sensor data acquisition and advanced data preprocessing techniques. Through real-time monitoring by sensors such as speed, temperature, pressure, vibration, and wear, the system can comprehensively acquire key information during the cutting process. After outlier removal, interpolation imputation, smoothing, and noise filtering, the quality and reliability of this data are effectively improved, reducing the impact of interference factors on control decisions. Furthermore, by normalizing data from different types of sensors, the uniformity of various data at the same scale is ensured, further improving the accuracy and robustness of subsequent model predictions. This method significantly enhances the real-time response capability and control accuracy of the pipe cutting system, ensuring the stability and efficiency of the cutting process, while reducing control failures caused by data inconsistency or noise interference, thus optimizing production efficiency and equipment performance.

[0022] In this embodiment, step two specifically includes: The standardized sensor dataset is sorted in chronological order to obtain time series data; Higher-order differencing is performed on the time series data to obtain differencing time series data. The specific steps of higher-order differencing are as follows: The original time series data is subjected to a first difference, which is to subtract the data of the previous time point from the data of each time point to obtain a new sequence. The new sequence is subjected to a second difference, which is to subtract the data of the previous time point from each time point in the new sequence to obtain the second difference data. Repeat the iteration until the difference between any two adjacent time points is less than a preset threshold to obtain differential time series data.

[0023] This step significantly improves the adaptability and accuracy of the pipe cutting system to time-series data through high-order differencing. By performing high-order differencing on standardized sensor data, long-term trends in the data are eliminated while retaining short-term fluctuations and nonlinear changes during the cutting process, enabling the model to more accurately capture real-time changing patterns. This processing method effectively enhances the model's sensitivity to instantaneous changes and improves its predictive ability for dynamic cutting conditions. Simultaneously, through stepwise differencing and preset threshold control, signal loss caused by excessive differencing is avoided, improving data validity and model prediction accuracy. This method provides the pipe cutting system with more precise real-time control parameters, ensuring high precision and stability in the cutting process, thereby improving production efficiency and reducing equipment failure rates.

[0024] In this embodiment, step three specifically includes: The differential time series data is segmented according to a preset time period length, with each time period containing a fixed number of data points; A sliding time window is used to progressively divide each time period into multiple time period subsets; Independent autoregressive modeling is performed on each time period subset to obtain the autoregressive prediction results for each time period subset. The specific autoregressive modeling steps are as follows: The order of autoregressive modeling is determined for a subset of time periods using the Akaike information content criterion, where the order is determined by the number of past time points used for prediction. The least squares method is used to estimate the autoregressive coefficients in the autoregressive model and to minimize the error between the predicted and actual values. The autoregressive model is used to calculate the predicted values ​​for future time points to obtain the autoregressive prediction results. The autoregressive prediction results of the subsets of data from each time period are weighted and averaged to obtain the global time series prediction result.

[0025] This step segmentes the differential time series data into preset time periods and uses a sliding time window to progressively divide each period, ensuring data continuity and stability. Within each time period, independent autoregressive modeling is performed, and the optimal model order is determined using the Akaike information content criterion, effectively avoiding overfitting and improving prediction accuracy. The least squares method is used to estimate the autoregressive coefficients and minimize prediction error, enabling the model to accurately capture patterns and trends in the time series. Finally, by weighted averaging the autoregressive prediction results from each time period, a global time series prediction result is synthesized, achieving accurate prediction of various control parameters for the pipe cutting system. This method improves control accuracy and adaptability during the cutting process, allowing for adjustments to control strategies based on real-time data, ensuring the stability and efficiency of the production process.

[0026] In this embodiment, step four specifically includes: Real-time acquisition of tube cutting image data during the cutting process using high-definition cameras or infrared sensors; The tube cutting image data includes tool wear, cutting edge, and material reflectivity; the tool wear includes the wear depth and morphological change characteristics of the tool. The cut edge includes the flatness and serration of the edge; the material reflectance is the surface smoothness of the material; The acquired tube-cutting image data is input into a ResNeXt network for feature extraction. The ResNeXt network extracts key information from the image in the following way: The tube-cutting image data is input into a ResNeXt network, and the tube-cutting image feature vector is obtained through grouped convolution and residual connections. The specific steps are as follows: ResNeXt divides the tube-cutting image data into multiple groups using grouped convolution technology. Each group is convolved using an independent convolution kernel, and multiple residual blocks are stacked. Each residual block contains a skip connection, multiple convolutional layers, and a ReLU activation function. The skip connection allows the input data to skip the convolutional layer and be passed directly to the pooling layer; the ReLU activation function performs a non-linear transformation, setting all negative values ​​to zero and keeping positive values ​​unchanged. Local features in the tube-cutting image are extracted by convolution operation. These local features include the depth of tool wear, morphological changes, and the shape of the cutting edge. Global features of the tube-cutting image are extracted through multiple convolutional and pooling layers to capture the overall structure in the tube-cutting image, including changes in the cutting surface quality and cutting angle; the pooling layer downsamples local features through max pooling operation. The global features are sorted according to the time series to obtain the feature vector of the tube-cutting image.

[0027] This step involves real-time acquisition of image data during the cutting process using a high-definition camera or infrared sensor. Combined with a ResNeXt network, efficient feature extraction is performed, accurately identifying key indicators such as tool wear, cutting edge quality, and material reflectivity. Through grouped convolution and residual connection techniques, the ResNeXt network effectively extracts not only local features (such as the depth and shape changes of tool wear) but also global features (such as cutting surface quality and cutting angle changes), giving the model greater flexibility and accuracy in image processing. This method significantly enhances the intelligent perception and control capabilities of the pipe cutting system during the cutting process, ensuring the stability and accuracy of cutting quality. Through real-time feedback and prediction, the system can adjust control parameters promptly, thereby reducing cutting errors, improving production efficiency, and extending equipment lifespan.

[0028] In this embodiment, step five specifically includes: The global time series prediction results and the feature vector of the tube-cutting image are concatenated along the feature dimension. The splicing step involves aligning the two vectors element by element, aligning them element by element along the feature dimension, and then connecting the two vectors to generate a long vector containing time series and tube-cutting image information, which serves as the final joint feature vector. The joint feature vector is input into the improved DeepAR model, and the posterior distribution of the prediction results is obtained through variational inference and Bayesian estimation. The specific steps are as follows: The prior distribution and marginal likelihood are obtained from the joint eigenvectors using the Laplace distribution; The posterior distribution is derived from the prior distribution and marginal likelihood using Bayes' theorem. Variational posterior distribution is obtained by using variational inference on the posterior distribution, wherein the variational inference involves selecting a Gaussian distribution as a family of distributions, and the family of distributions is used to approximate the variational posterior distribution. Based on the variational posterior distribution, Bayesian estimation is used to update the value of each prediction parameter to obtain the posterior distribution of the prediction result. The Bayesian estimation steps are as follows: Calculate the posterior expectation of each parameter in the variational posterior distribution to obtain the predicted value of each parameter; Calculate the variance or standard deviation of the variational posterior distribution, provide the confidence interval for the predicted value of each parameter, and obtain the posterior distribution of the prediction result. The confidence interval represents the fluctuation range of the predicted value.

[0029] This step concatenates the global time-series prediction results and the feature vectors of the pipe-cutting image along the feature dimension to generate a joint feature vector, fully integrating time-series and image data, thereby improving the model's predictive ability. After the joint feature vector is input into the improved DeepAR model, variational inference and Bayesian estimation enable the system to effectively quantify the uncertainty of the prediction results, providing predicted values ​​and confidence intervals for each control parameter. This method not only enhances the adaptability of the pipe-cutting system to dynamic changes but also provides more accurate prediction results through Bayesian estimation and variational inference, enabling the system to make more robust control decisions in environments with high uncertainty, thus optimizing the accuracy and stability of the cutting process.

[0030] In this embodiment, step six specifically includes: Based on the posterior distribution of the prediction results, the most probable value and the corresponding confidence interval of each prediction control parameter are extracted. The prediction control parameters include cutting speed, feed speed, feed angle, cutting pressure and cutting path. The most probable value is the expected value of the corresponding posterior distribution. Adjust the tool feed rate according to the predicted feed rate to increase the material removal rate during the cutting process; By adjusting various predictive control parameters, corresponding control commands are generated, including motor control commands and tool control commands.

[0031] In this embodiment, step seven specifically includes: The actual operating data is obtained by real-time monitoring of various data during the cutting process using sensors, including cutting speed, feed rate, cutting pressure, tool wear, and temperature parameters. The actual operating data monitored are compared with the previously predicted control parameters, and the difference is calculated. If the difference exceeds a preset threshold, the control parameters are adjusted according to the magnitude of the deviation. The adjustment involves proportionally scaling the cutting speed, feed rate, cutting pressure, and tool angle.

[0032] Example 1: To verify the feasibility of this invention in practice, it was applied to a pipe-cutting production line in a precision machining plant. The study investigated how to utilize a real-time control method based on a digital twin-based bus-type pipe-cutting system to optimize control parameters during the cutting process, thereby improving cutting quality, increasing production efficiency, and reducing equipment wear. On this plant's production line, the pipe-cutting system is primarily used for the precision cutting of pipes, with high product quality requirements. However, due to the numerous variable factors in the cutting process, traditional control methods based on fixed rules are ill-suited to adapt to these variations, often resulting in large cutting errors and severe tool wear.

[0033] To address this issue, the factory introduced a real-time control method for a bus-type pipe cutting system based on digital twins. By integrating multi-dimensional sensors and an image data acquisition system, it achieved real-time monitoring and adjustment of the pipe cutting process. First, speed sensors monitor the movement speed of the cutting tool and its feed rate across the pipe surface in real time, ensuring that the relative position of the tool and workpiece always meets predetermined requirements during the cutting process. Simultaneously, temperature sensors acquire temperature data of the cutting area and the cutting tool, preventing tool damage or reduced cutting quality due to overheating. Pressure sensors detect the cutting force or pressure in real time during the cutting process, ensuring that the applied force remains within a safe range, thereby reducing fluctuations in cutting quality and tool wear.

[0034] The vibration status of the cutting tool is also monitored by vibration sensors. Vibration data provides direct feedback on cutting stability, and combined with tool wear status, it can further determine the usage condition of the cutting tool. If the vibration amplitude is too large or the tool wear is severe, the system can automatically adjust the cutting parameters or trigger a tool replacement reminder. Furthermore, all sensor data undergoes real-time processing, including noise reduction, filtering, and interpolation, to form a standardized sensor dataset. These processing steps ensure data accuracy and reliability, providing a high-quality data foundation for subsequent real-time control. Regarding image data, high-definition cameras and infrared sensors are used to acquire real-time images of the cut tube during the cutting process. Using a ResNeXt network to extract features from the images, the system can extract the depth and morphological changes of tool wear, as well as the smoothness and serration of the cutting edge. Combining time-series data, a joint feature vector is generated by concatenating the image feature vector with the global time-series prediction results. This joint feature vector is then input into an improved DeepAR model for time-series prediction, ultimately generating the predicted control parameters.

[0035] This system enables the production line to calculate and adjust control parameters in real time, such as cutting speed, feed rate, and cutting path. When the deviation between the actual data and the predicted results during the cutting process exceeds a preset threshold, the system automatically corrects it. For example, in a cutting process, the system detects that the actual cutting speed and feed rate deviate from the predicted values ​​by 3% and 5%, respectively. By calculating the deviation and adjusting the control parameters, the system ensures that the cutting quality remains stable within the predetermined range, avoiding errors caused by manual adjustments in traditional methods.

[0036] The system's performance was validated during production, particularly in complex cutting processes. Traditional cutting methods often resulted in significant cutting errors, while the real-time control method of this invention significantly improved cutting accuracy. The table below shows a comparison of data before and after the cutting process in a batch of high-precision pipes.

[0037] Table 1 Performance Comparison of Control Methods for Pipe Cutting Systems

[0038] As shown in Table 1, traditional rule-based control methods and simple regression model-based methods exhibit significant errors in cutting path deviation, at ±0.6mm and ±0.4mm respectively. This can lead to failure to meet product quality standards in high-precision cutting processes. The method of this invention, through multi-dimensional data fusion and dynamic adjustment, reduces the cutting path deviation to ±0.1mm, significantly improving cutting accuracy. Traditional methods perform poorly in tool wear control, resulting in significant tool wear, particularly under rule-based control and simple regression model conditions, reaching 0.25mm and 0.20mm respectively. However, through the real-time control and feedback correction mechanism of the method of this invention, tool wear is significantly reduced to 0.12mm. This means that the method of this invention not only improves cutting quality but also extends tool life. Traditional methods, especially rule-based control methods, struggle to maintain stable control under dynamically changing cutting conditions. While traditional methods based on simple regression models improve stability to some extent, they still lack adaptability to real-time changes. In contrast, the method of this invention utilizes real-time feedback of multimodal data and Bayesian estimation to quantify uncertainty, maintaining excellent control stability under complex working conditions and ensuring that each control parameter is precisely adjusted throughout the cutting process. This method demonstrates superior performance in cutting accuracy, tool wear control, and system stability, improving the overall performance of the pipe cutting system, reducing the failure rate during production, and optimizing cutting quality while meeting high precision and high reliability requirements, showcasing significant technical advantages.

[0039] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A real-time control method for a bus-type pipe cutting system based on digital twins, characterized in that, Includes the following steps: Step 1: Collect multidimensional sensor data and preprocess the multidimensional sensor data to obtain a standardized sensor dataset; Step 2: Perform high-order differencing based on the standardized sensor dataset to obtain differential time series data; Step 3: Divide the differential time series data into segments according to a preset time period, perform independent autoregressive modeling in each time period, and fuse the autoregressive prediction results of each time period to obtain the global time series prediction result. Step 4: Real-time acquisition of pipe-cutting image data during the cutting process, and inputting the pipe-cutting image data into the ResNeXt network to obtain the pipe-cutting image feature vector through grouped convolution and residual connections; Step 5: Fuse the global time series prediction results and the tube-cutting image feature vector to generate a joint feature vector, and input the joint feature vector into the improved DeepAR model. Through variational inference and Bayesian estimation, the posterior distribution of the prediction results is obtained. Step 6: Based on the posterior distribution of the predicted results, calculate and adjust the control parameters in real time, and generate corresponding control commands; Step 7: Monitor the difference between the actual operating data and the predicted results during the cutting process. When the deviation exceeds the preset threshold, adjust the control parameters.

2. The real-time control method for a bus-type pipe cutting system based on digital twins according to claim 1, characterized in that, Step one specifically involves: The speed of the cutting tool and the feed rate as it moves across the pipe surface are monitored in real time using a speed sensor. Temperature sensors are used to acquire temperature data of the cutting area and the cutting tool. The cutting force or pressure during the cutting process is detected by a pressure sensor; Vibration sensors are used to collect the vibration status of the cutting tool and the wear status of the cutting tool. Remove outliers from the multidimensional sensor data and fill in the missing values ​​in the sensor data using an average interpolation method; A low-pass filter is used to smooth the multidimensional sensor data and remove high-frequency noise; And use a high-pass filter to remove low-frequency noise; Different types of data in the multidimensional sensor data are normalized separately to obtain a standardized sensor dataset.

3. The real-time control method for a bus-type pipe cutting system based on digital twins according to claim 1, characterized in that, Step two specifically involves: The standardized sensor dataset is sorted in chronological order to obtain time series data; Higher-order differencing is performed on the time series data to obtain differencing time series data. The specific steps of higher-order differencing are as follows: The original time series data is subjected to a first difference, which is to subtract the data of the previous time point from the data of each time point to obtain a new sequence. The new sequence is subjected to a second difference, which is to subtract the data of the previous time point from each time point in the new sequence to obtain the second difference data. Repeat the iteration until the difference between any two adjacent time points is less than a preset threshold to obtain differential time series data.

4. The real-time control method for a bus-type pipe cutting system based on digital twins according to claim 1, characterized in that, Step three specifically involves: The differential time series data is segmented according to a preset time period length, with each time period containing a fixed number of data points; A sliding time window is used to progressively divide each time period into multiple time period subsets; Independent autoregressive modeling is performed on each time period subset to obtain the autoregressive prediction results for each time period subset. The specific autoregressive modeling steps are as follows: The order of autoregressive modeling is determined for a subset of time periods using the Akaike information content criterion, where the order is determined by the number of past time points used for prediction. The least squares method is used to estimate the autoregressive coefficients in the autoregressive model and to minimize the error between the predicted and actual values. The autoregressive model is used to calculate the predicted values ​​for future time points to obtain the autoregressive prediction results. The autoregressive prediction results of the subsets of data from each time period are weighted and averaged to obtain the global time series prediction result.

5. The real-time control method for a bus-type pipe cutting system based on digital twins according to claim 1, characterized in that, Step four specifically involves: Real-time acquisition of tube cutting image data during the cutting process using high-definition cameras or infrared sensors; The tube cutting image data includes tool wear, cutting edge, and material reflectivity; the tool wear includes the wear depth and morphological change characteristics of the tool. The cut edge includes the flatness and serration of the edge; the material reflectance is the surface smoothness of the material; The acquired tube-cutting image data is input into a ResNeXt network for feature extraction. The ResNeXt network extracts key information from the image in the following way: The tube-cutting image data is input into a ResNeXt network, and the tube-cutting image feature vector is obtained through grouped convolution and residual connections. The specific steps are as follows: ResNeXt divides the tube-cutting image data into multiple groups using grouped convolution technology. Each group is convolved using an independent convolution kernel, and multiple residual blocks are stacked. Each residual block contains a skip connection, multiple convolutional layers, and a ReLU activation function. The skip connection allows the input data to skip the convolutional layer and be passed directly to the pooling layer; the ReLU activation function performs a non-linear transformation, setting all negative values ​​to zero and keeping positive values ​​unchanged. Local features in the tube-cutting image are extracted by convolution operation. These local features include the depth of tool wear, morphological changes, and the shape of the cutting edge. Global features of the tube-cutting image are extracted through multiple convolutional and pooling layers to capture the overall structure in the tube-cutting image, including changes in the cutting surface quality and cutting angle; the pooling layer downsamples local features through max pooling operation. The global features are sorted according to the time series to obtain the feature vector of the tube-cutting image.

6. The real-time control method for a bus-type pipe cutting system based on digital twins according to claim 1, characterized in that, Step five specifically involves: The global time series prediction results and the feature vector of the tube-cutting image are concatenated along the feature dimension. The splicing step involves aligning the two vectors element by element, aligning them element by element along the feature dimension, and then connecting the two vectors to generate a long vector containing time series and tube-cutting image information, which serves as the final joint feature vector. The joint feature vector is input into the improved DeepAR model, and the posterior distribution of the prediction results is obtained through variational inference and Bayesian estimation. The specific steps are as follows: The prior distribution and marginal likelihood are obtained from the joint eigenvectors using the Laplace distribution; The posterior distribution is derived from the prior distribution and marginal likelihood using Bayes' theorem. Variational posterior distribution is obtained by using variational inference on the posterior distribution, wherein the variational inference involves selecting a Gaussian distribution as a family of distributions, and the family of distributions is used to approximate the variational posterior distribution. Based on the variational posterior distribution, Bayesian estimation is used to update the value of each prediction parameter to obtain the posterior distribution of the prediction result. The Bayesian estimation steps are as follows: Calculate the posterior expectation of each parameter in the variational posterior distribution to obtain the predicted value of each parameter; Calculate the variance or standard deviation of the variational posterior distribution, provide the confidence interval for the predicted value of each parameter, and obtain the posterior distribution of the prediction result. The confidence interval represents the fluctuation range of the predicted value.

7. The real-time control method for a bus-type pipe cutting system based on digital twins according to claim 1, characterized in that, Step six specifically involves: Based on the posterior distribution of the prediction results, the most probable value and the corresponding confidence interval of each prediction control parameter are extracted. The prediction control parameters include cutting speed, feed speed, feed angle, cutting pressure and cutting path. The most probable value is the expected value of the corresponding posterior distribution. Adjust the tool feed rate according to the predicted feed rate to increase the material removal rate during the cutting process; By adjusting various predictive control parameters, corresponding control commands are generated, including motor control commands and tool control commands.

8. The real-time control method for a bus-type pipe cutting system based on digital twins according to claim 1, characterized in that, Step seven specifically involves: The actual operating data is obtained by real-time monitoring of various data during the cutting process using sensors, including cutting speed, feed rate, cutting pressure, tool wear, and temperature parameters. The actual operating data monitored are compared with the previously predicted control parameters, and the difference is calculated. If the difference exceeds a preset threshold, the control parameters are adjusted according to the magnitude of the deviation. The adjustment involves proportionally scaling the cutting speed, feed rate, cutting pressure, and tool angle.

Images