A high-precision plasma cutting method and equipment

Abstract

This invention discloses a high-precision plasma cutting method, comprising the following steps: S1, acquiring real-time voltage information, connecting the feedback circuit of the cutting torch to an FPGA, and obtaining the real-time voltage value a of the cutting torch. i S2, Obtain the desired voltage value a0 in the MCU; S3, Compare the real-time voltage value with the desired voltage value; S4, Compare the real-time voltage value with the desired voltage value and calculate the error value; S5, Based on the error value, calculate the control quantity through proportional, integral, and derivative operations; S6, Use the calculated control quantity as the basis for adjusting the power output, and put the control quantity into the desired voltage value to obtain the required input voltage value; This invention adjusts the voltage of the cutting torch through a PID algorithm and proximity method, so that it can be free from the influence of environmental factors, and adjust the desired voltage value and the real-time voltage value to the absolute value range of 2. This threshold does not affect the actual cutting accuracy and speed of the cutting torch.

Description

Technical Field

[0001] This invention relates to the field of plasma cutting, and in particular to a high-precision plasma cutting method and equipment. Background Technology

[0002] Variations in the ambient temperature of plasma cutting equipment can affect the power supply and the cutting torch. In high-temperature environments, the performance of the electronic components inside the power supply can change; for example, the conduction characteristics of semiconductor devices may alter, potentially leading to a decrease in the power supply's output voltage. Simultaneously, the cutting torch may experience performance degradation due to poor heat dissipation in high-temperature environments, affecting voltage stability.

[0003] In low-temperature environments, the performance of semiconductor devices (such as transistors and diodes) in a power supply is significantly affected. Taking a transistor as an example, its base-emitter voltage increases as the temperature decreases, the power supply's output current decreases, and the output voltage increases under constant load, thus affecting the normal operating voltage of the cutting torch. Since voltage has a crucial impact on the cutting torch, if the output voltage does not match the desired voltage, the cutting torch will be affected in various aspects such as cutting speed and accuracy. Summary of the Invention

[0004] In order to solve the above-mentioned technical problems, the present invention provides a high-precision plasma cutting method and equipment.

[0005] The technical solution of this invention is implemented as follows:

[0006] A high-precision plasma cutting method includes the following steps:

[0007] S1, collect real-time voltage information, connect the torch feedback circuit to the FPGA, and obtain the torch real-time voltage value a. i ;

[0008] S2, obtain the desired voltage value a0 within the MCU;

[0009] S3 compares the real-time voltage value with the expected voltage value;

[0010] S4 compares the real-time voltage value with the expected voltage value and calculates the error value;

[0011] S5. Based on the error value, the control quantity is calculated through proportional, integral, and derivative operations. S6. The calculated control quantity is used as the basis for adjusting the power supply output. The control quantity is then placed into the desired voltage value to obtain the required input voltage value.

[0012] Preferably, in step S5, the proportional calculation formula is e = V 期望 -V 实时 u p =Kp ×e; where e is the error value, u p To control the quantity, K p As a scaling factor, K is first set during the initial debugging phase. p Set K to 0.1 and observe the system response. If the real-time voltage converges to the desired voltage too slowly, gradually increase K. p The absolute value is calculated until the system can respond quickly to the error without significant oscillation.

[0013] Preferably, in step S5, the formula for the integral is: Among them, K i The integral coefficient is t, and time is t. The integral term will continuously accumulate error. As time goes by, even if the error is small, the integral control quantity will gradually increase, thereby pushing the real-time voltage closer to the desired voltage.

[0014] Preferably, in step S5, the formula for the differential is u. d =K d ×d e / d t During the initial debugging phase, first set K. d Set the value to 0.1 and observe the system response. If the real-time voltage converges to the desired voltage too slowly, gradually increase the absolute value of Kd until the system can quickly respond to the error without significant oscillation.

[0015] Preferably, in step S6, the control quantity is the sum of the proportional, integral, and derivative control quantities, u = u p +u i +u d By obtaining the total control quantity u and the real-time measured voltage value V, the ratio k between the control quantity and the real-time voltage value V can be determined, k = V / u. Finally, the adjusted input voltage value V1 can be obtained, V1 = V + k × u.

[0016] A high-precision plasma cutting device includes a plasma cutting torch for implementing the aforementioned high-precision plasma cutting method.

[0017] This invention uses a PID algorithm and proximity method to adjust the voltage of the cutting torch, so that it can be free from the influence of environmental factors and adjust the desired voltage value and the real-time voltage value to a range of absolute values ​​of 2. This threshold does not affect the actual cutting accuracy and speed of the cutting torch. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of a high-precision plasma cutting method according to the present invention. Detailed Implementation

[0019] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, exemplary embodiments will be described in detail below, examples of which are illustrated in the accompanying drawings. In the following description relating to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of methods and systems consistent with some aspects of this application as detailed in the appended claims.

[0020] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0021] The following detailed description of the specific implementation methods, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided in detail.

[0022] Example 1

[0023] This invention provides a high-precision plasma cutting method, comprising the following steps:

[0024] S1, collect real-time voltage information, connect the torch feedback circuit to the FPGA, and obtain the torch real-time voltage value a. i ;

[0025] S2, obtain the desired voltage value a0 within the MCU;

[0026] S3, compare the real-time voltage value with the expected voltage value; in practical applications, due to the influence of seasonality and regionality on temperature and humidity, the difference between the real-time voltage value and the expected voltage value is large, requiring real-time adjustment. To address the influence of seasonality and regionality on temperature and humidity, step S3 further includes the following sub-steps:

[0027] S31, Create a table to record historical data, including real-time voltage values ​​and expected voltage values;

[0028] S32, From the table, we can see the linear relationship between the real-time voltage value and the expected voltage value, K0;

[0029] S33, the voltage is adjusted for the first time. Through K0, the output voltage value a1 is obtained as: desired voltage value a0 × linear relationship value K0;

[0030] S34. Compare a1 and a0. If the comparison result is within the absolute value of 20, skip steps S4 and S5. If the comparison result exceeds the absolute value of 20, calculate the error value and control quantity in steps S4 and S5. However, K0 will change with the season and region. Therefore, the linear relationship value is only the average of the first 10 days of the table and the 20 days before and after the same month and day in previous years. This is the linear relationship value K0 at this time.

[0031] S4 compares the real-time voltage value with the expected voltage value and calculates the error value;

[0032] S5 calculates the control quantity based on the error value through the proportional, integral, and derivative (PID) operations.

[0033] S6, use the calculated control quantity as the basis for adjusting the power supply output, put the control quantity into the desired voltage value, and thus obtain the required input voltage value;

[0034] Step S6 further includes the following sub-steps:

[0035] S61, acquire the real-time voltage value a again. t , will a t Compare with the expected voltage value;

[0036] S62, by comparing the results, for a t Perform coarse adjustment using the proximity method, adjusting a each time. t Move 5 units closer to a0, until a t The comparison result with a0 is within 20 in absolute value;

[0037] S63, for a t Make fine adjustments, adjusting by ±2 units each time, moving closer to a0, until the comparison result value is within the absolute value of 2.

[0038] Preferably, in step S5, the proportional calculation formula is e = V 期望 -V 实时 u p =K p ×e; where e is the error value, u p To control the quantity, K p As a scaling factor, K is first set during the initial debugging phase. p Set K to 0.1 and observe the system response. If the real-time voltage converges to the desired voltage too slowly, gradually increase K. p The absolute value is calculated until the system can respond quickly to the error without significant oscillation.

[0039] Preferably, in step S5, the formula for the integral is: Among them, Ki The integral coefficient is t, and time is t. The integral term will continuously accumulate error. As time goes by, even if the error is small, the integral control quantity will gradually increase, thereby pushing the real-time voltage closer to the desired voltage.

[0040] Preferably, in step S5, the formula for the differential is u. d =K d ×d e / d t During the initial debugging phase, first set K. d Set K to 0.1 and observe the system response. If the real-time voltage converges to the desired voltage too slowly, gradually increase K. d The absolute value is calculated until the system can respond quickly to the error without significant oscillation.

[0041] Preferably, in step S6, the control quantity is the sum of the proportional, integral, and derivative control quantities, u = u p +u i +u d By obtaining the total control quantity u and the real-time measured voltage value V, the ratio k between the control quantity and the real-time voltage value V can be determined, k = V / u. Finally, the adjusted input voltage value V1 can be obtained, V1 = V + k × u.

[0042] Preferably, the feedback circuit described in step S1 is an important component for transmitting the nozzle voltage signal to the FPGA for acquisition and processing, including,

[0043] Sensors are used to detect and measure changes in nozzle voltage;

[0044] In feedback circuits, amplifiers are typically used to amplify the voltage signal obtained from the sensor and match the sensor's output. Filters are used to remove noise and interference from the sensor output, ensuring that the accurate voltage signal is transmitted to the FPGA. Analog-to-digital converters (ADCs) are key components in feedback circuits, used to convert analog voltage signals into digital AD values; ADCs discretize continuous analog voltage signals.

[0045] A high-precision plasma cutting device includes a plasma cutting torch for implementing the aforementioned high-precision plasma cutting method.

[0046] Example 2

[0047] S1, Data Acquisition: Connect the torch feedback circuit to the FPGA to obtain the real-time torch voltage value. At the same time, install temperature and humidity sensors, air pressure sensors, etc. to collect environmental data. In addition, it is also necessary to collect parameters such as torch current value and plasma gas flow rate to provide more input features for the neural network. Continuous data over a period of time is collected to form time series data so that the neural network can better learn dynamic changes.

[0048] S2, Data labeling and organization, involves labeling the collected data, including the corresponding expected voltage value and the evaluation level of the cutting effect.

[0049] Data is organized and normalized to bring data of different dimensions into a similar numerical range, which facilitates the training of neural networks.

[0050] S3, the neural network model construction stage, can select a deep neural network with good nonlinear fitting ability, such as a combination of multilayer perceptron (MLP), convolutional neural network (CNN) and long short-term memory network (LST M), etc. The input layer receives parameters such as real-time voltage value, temperature and humidity, air pressure, current value, and plasma gas flow rate, and the output layer is the desired voltage value.

[0051] S4, Model Training: The prepared data is divided into training, validation, and test sets. The neural network is trained using the training set, and the weights and biases of the network are continuously adjusted through the backpropagation algorithm to minimize the error between the predicted and actual values.

[0052] During training, use a validation set to monitor model performance and prevent overfitting. When the model's performance on the validation set no longer improves, training can be stopped, for example, by using deep learning frameworks such as TensorFlow or PyTorch to build and train the neural network model.

[0053] S5, the application model stage, involves inputting various parameters collected in real time into the trained neural network model during the actual cutting process. The model will then output the dynamic expected voltage value.

[0054] Because neural networks can learn the complex relationships between different environmental parameters and operating states and the desired voltage, they can adapt more accurately to seasonal and regional influences.

[0055] In addition to estimating the expected output voltage value, neural networks can also estimate the uncertainty of the output. For example, methods such as Monte Carlo dropout can be used to run the neural network multiple times and observe the range of output variation to understand the degree of uncertainty in the prediction.

[0056] If there is significant uncertainty, a more conservative adjustment strategy can be adopted, or additional monitoring and adjustments can be made.

[0057] S6, Voltage Comparison and Adjustment Stage: Intelligent comparison and feedback, comparing the real-time voltage value with the expected voltage value predicted by the neural network, and calculating the error.

[0058] Error information is fed back to the neural network so that the network can take this error into account in subsequent predictions, thereby further improving the accuracy of the predictions.

[0059] Adaptive adjustment strategy:

[0060] An adaptive adjustment strategy is formulated based on the magnitude of the error and the uncertainty estimation of the neural network. If the error is small and the uncertainty is low, a smaller adjustment step size can be used; if the error is large or the uncertainty is high, more aggressive adjustment measures can be taken, such as increasing the frequency or magnitude of the adjustment.

[0061] S7, the power output adjustment and feedback stage, is an intelligent power control system that uses the expected voltage value predicted by the neural network as the basis for adjusting the power output. It communicates with the intelligent power controller through a digital interface to precisely adjust the power output. The power controller can dynamically adjust the output based on parameters such as real-time feedback voltage and current values ​​to ensure that the actual output voltage is close to the expected voltage.

[0062] Closed-loop feedback and model updates: Establish a strict closed-loop feedback system to continuously monitor various parameters and cutting effects during the actual cutting process.

[0063] If the actual cutting effect deviates significantly from the expectation, this new data can be added to the training set to update the neural network online or offline, thereby continuously optimizing the model's performance.

[0064] Preferably, the closed-loop feedback system can continuously monitor various key parameters during the actual cutting process, such as the real-time voltage value of the torch, temperature and humidity, gas pressure, current value, plasma gas flow rate, and actual cutting effect. This real-time monitoring enables the system to promptly detect deviations from the desired state and take rapid adjustment measures.

[0065] For example, when the real-time voltage value deviates significantly from the expected voltage value predicted by the neural network, the system can immediately trigger the power output adjustment mechanism, which uses an intelligent power controller to quickly change the output voltage to bring the actual voltage closer to the expected voltage.

[0066] The closed-loop feedback system effectively reduces the impact of external interference and internal uncertainties on the cutting process by continuously comparing and adjusting the actual output with the desired target. This helps improve the system's stability and reliability, ensuring good cutting performance under different working environments and conditions.

[0067] For example, even under conditions of significant seasonal and regional temperature and humidity variations, the closed-loop feedback system can adjust system parameters in a timely manner based on real-time monitoring data to ensure the stability of the cutting process.

[0068] During plasma cutting, various factors may change dynamically over time, such as material thermal deformation, plasma gas consumption, and equipment wear. A closed-loop feedback system can sense these changes in real time and adjust system parameters accordingly, ensuring the system remains in optimal operating condition.

[0069] For example, as cutting time increases, the torch temperature may rise, potentially affecting torch performance and voltage output. A closed-loop feedback system can monitor torch temperature changes and update the neural network online by adding new data to the training set in real-time during the actual cutting process. This method allows the model to quickly adapt to changes in real-world conditions, improving its real-time performance and accuracy.

[0070] For example, when the system detects a significant deviation between the actual cutting effect and the expected result, it can immediately add the current parameters (including real-time voltage, temperature, humidity, air pressure, current, etc.) and the actual cutting effect as new data points to the training set. Then, optimization algorithms such as mini-batch stochastic gradient descent are used to update the neural network online, enabling the model to adjust more quickly to adapt to the new situation.

[0071] The frequency of online updates can be adjusted according to the actual situation. If the changes during the cutting process are frequent, a higher update frequency can be set; if the changes are relatively slow, the update frequency can be appropriately reduced to avoid over-updating and causing model instability.

[0072] S8 involves collecting a large amount of new data at regular intervals and then retraining the neural network. This method can more comprehensively consider the long-term trends of the system and improve the model's generalization ability. For example, new data can be collected daily or weekly, and the neural network can be updated offline during off-peak hours. During the update process, more training techniques, such as data augmentation and regularization, can be used to improve the model's performance and stability.

[0073] Offline updates can be combined with online updates to form a hybrid update strategy. For example, online updates can be performed during routine slicing to quickly adapt to short-term changes, while offline updates can be performed during regular maintenance to comprehensively optimize model performance.

[0074] S9 involves selecting and filtering newly added data to ensure its quality and validity. Data can be filtered according to certain criteria, such as removing outliers and selecting representative data points. For example, data quality indicators can be set, such as data completeness, accuracy, and consistency. Only data that meets these indicators will be added to the training set for model updates. Furthermore, data can be weighted according to its importance, giving more significant data a greater impact on model updates. S9 then evaluates and validates the updated model to ensure improved performance. A validation set or test set can be used to evaluate the model, comparing performance metrics before and after the update, such as accuracy and mean squared error.

[0075] For example, if the updated model performs significantly better on the validation set than the original model, the model update is effective; if the performance does not improve significantly or even declines, it may be necessary to adjust the update strategy or reconsider the model's architecture and parameters.

[0076] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A high-precision plasma cutting method, characterized in that, Includes the following steps: S1, Data Acquisition: The feedback circuit of the cutting torch is connected to the FPGA to obtain the real-time voltage value of the cutting torch. At the same time, temperature and humidity sensors and air pressure sensors are installed to collect environmental data, as well as the current value and plasma gas flow parameters of the cutting torch, to provide more input features for the neural network. Continuous data over a period of time is collected to form time series data, so that the neural network can better learn dynamic changes. S2, Data labeling and organization, involves labeling the collected data, including the corresponding expected voltage value and the evaluation level of the cutting effect; Specifically, the data is organized and normalized to bring data of different dimensions into a similar numerical range, which facilitates the training of neural networks; S3, the neural network model construction stage, selects a deep neural network with good nonlinear fitting ability, specifically a combination of multilayer perceptron (MLP), convolutional neural network (CNN) and long short-term memory network (LSTM). The input layer receives parameters such as real-time voltage value, temperature and humidity, air pressure, current value and plasma gas flow rate, and the output layer is the desired voltage value. S4, Model Training: The prepared data is divided into training set, validation set and test set. The neural network is trained using the training set. The weights and biases of the network are continuously adjusted through the backpropagation algorithm to minimize the error between the predicted value and the actual value. During training, the model's performance is monitored using a validation set to prevent overfitting. Training is stopped when the model's performance on the validation set no longer improves. S5, the application model stage: During the actual cutting process, various parameters collected in real time are input into the trained neural network model, and the model will output a dynamic expected voltage value. Because neural networks can learn the complex relationships between different environmental parameters and operating states and the desired voltage value, they can adapt more accurately to seasonal and regional influences. In addition to outputting the expected voltage value, the application model stage can also estimate the uncertainty of the output. By using the Monte Carlo dropout method, the neural network is run multiple times and the range of output changes is observed to understand the degree of uncertainty in the prediction. If the uncertainty is large, a more conservative adjustment strategy is adopted, or additional monitoring and adjustments are performed. S6, Voltage Comparison and Adjustment Stage: Intelligent comparison and feedback compares the real-time voltage value with the expected voltage value predicted by the neural network, calculates the error, and feeds the error information back to the neural network so that the network can take this error into account in subsequent predictions, further improving the accuracy of the prediction; based on the magnitude of the error and the uncertainty estimate of the neural network, an adaptive adjustment strategy is formulated. If the error is small and the uncertainty is low, a small adjustment step size is used; if the error is large or the uncertainty is high, the adjustment frequency or magnitude is increased. S7, power output adjustment and feedback stage, intelligent power control, uses the expected voltage value predicted by the neural network as the basis for adjusting the power output, communicates with the intelligent power controller through the digital interface, and precisely adjusts the power output; The power controller dynamically adjusts the output based on real-time feedback of voltage and current values ​​to ensure that the actual output voltage is close to the desired voltage value. Closed-loop feedback and model update: Establish a strict closed-loop feedback system to continuously monitor various parameters and cutting effects during the actual cutting process; If the actual cutting effect deviates significantly from the expectation, this new data is added to the training set, and the neural network is updated online or offline to continuously optimize the model's performance.

2. The high-precision plasma cutting method according to claim 1, characterized in that, The closed-loop feedback system can continuously monitor various key parameters during the actual cutting process, such as the real-time voltage value of the torch, temperature and humidity, air pressure, current value, plasma gas flow rate, and actual cutting effect. When the real-time voltage value deviates significantly from the expected voltage value predicted by the neural network, the system immediately triggers the power output adjustment mechanism, which quickly changes the output voltage through the intelligent power controller to make the actual voltage value approach the expected voltage value.

3. The high-precision plasma cutting method according to claim 1, characterized in that, The online update specifically involves adding new data to the training set in real time during the actual cutting process and updating the neural network. When the system detects a significant deviation between the actual cutting effect and the expected result, it immediately adds the current real-time voltage, temperature, humidity, air pressure, current parameters, and the actual cutting effect as new data points to the training set. Then, it uses a mini-batch stochastic gradient descent optimization algorithm to update the neural network online, enabling the model to adjust more quickly to adapt to new situations.

4. The high-precision plasma cutting method according to claim 1, characterized in that, The method further includes step S8, which involves collecting a large amount of new data within a certain time interval and then retraining the neural network. The certain time interval is set to be daily or weekly, and new data is collected once a day. The neural network is also updated offline during non-working hours.

5. The high-precision plasma cutting method according to claim 1, characterized in that, The offline update can be combined with the online update to form a hybrid update strategy. Online updates are performed during routine data splitting to quickly adapt to short-term changes, while offline updates are performed during regular maintenance to comprehensively optimize model performance.

6. The high-precision plasma cutting method according to claim 1, characterized in that, The method further includes step S9, which involves selecting and filtering newly added data to ensure data quality and validity, removing outliers, selecting representative data points, and setting data quality indicators, including data completeness, accuracy, and consistency. Only data that meets these indicators will be added to the training set for model updates and will be weighted according to the importance of the data so that important data has a greater impact on model updates.

7. The high-precision plasma cutting method according to claim 6, characterized in that, The method further includes step S10, which evaluates and validates the updated model to ensure that the model's performance has been improved. The model is evaluated using a validation set or a test set, and the performance metrics of the model before and after the update are compared. If the updated model performs significantly better than the model before the update on the validation set, it indicates that the model update is effective. If the performance does not improve significantly or even declines, the update strategy needs to be adjusted or the model's architecture and parameters need to be reconsidered.

8. A high-precision plasma cutting device, characterized in that, The plasma cutting torch includes the plasma cutting method described in any one of claims 1-7.

Images