Electric spark discharge machine control method and system based on industrial data processing

Abstract

The invention provides an electric spark discharge machining machine control method and system based on industrial data processing, and relates to the technical field of industrial data processing.The method comprises the steps that in the movement process of an electric spark machining machine, working conditions are captured, and high-frequency real-time sensing data flow is output; receiving a high-frequency real-time sensing data stream through a redundant dual-channel RS485 bus, and synchronously calling a processing position time sequence slice; isomorphic processing scene matching is carried out to obtain a historical control parameter sequence; outputting a multi-modal real-time working condition feature tensor; carrying out discharge parameter increment prediction, and outputting an updated pulse parameter adjustment vector; sending the parameters to a PLC (Programmable Logic Controller) to drive a discharge module of the processing machine to execute smooth parameter adjustment transition; and iteratively executing closed-loop prediction parameter adjustment. The technical problems that in the prior art, an electric spark machining machine often depends on manual adjustment and experience to select discharge parameters, so that the precision and consistency of the machining process are difficult to guarantee, the debugging period is long, and the production efficiency is low are solved.

Description

Technical Field

[0001] This invention relates to the field of industrial data processing technology, and specifically to a control method and system for electrical discharge machining (EDM) machines based on industrial data processing. Background Technology

[0002] Electrical discharge machining (EDM) is a specialized machining method widely used for processing hard and brittle materials, complex-shaped parts, and precision molds. However, EDM presents several challenges, particularly in the precise control and adjustment of pulse discharge parameters during the machining process. Traditional EDM machines often rely on manual adjustment and experience to select discharge parameters. Operators need to manually adjust parameters such as current and pulse frequency based on factors such as the material, shape, and size of different workpieces. The control of details during parameter adjustment is often influenced by the operator's experience, leading to unstable machining quality. Frequent manual intervention is not only time-consuming but also prone to adjustment errors. This problem makes it difficult to guarantee the accuracy and consistency of the machining process, resulting in long setup cycles and low production efficiency. Summary of the Invention

[0003] This application provides a control method and system for electrical discharge machining (EDM) machines based on industrial data processing, aiming to solve the technical problems of existing EDM machines often relying on manual adjustment and experience to select discharge parameters, which makes it difficult to guarantee the accuracy and consistency of the machining process, resulting in long debugging cycles and low production efficiency.

[0004] The first aspect disclosed in this application provides a control method for an electrical discharge machining (EDM) machine based on industrial data processing. The method includes: during the movement of the EDM machine, synchronously capturing operating conditions using a multi-physics field sensor array and outputting a high-frequency real-time sensor data stream; receiving the high-frequency real-time sensor data stream via a redundant dual-channel RS485 bus and synchronously retrieving a machining position time-series slice from the closed-loop position feedback module of the CNC system; performing isomorphic machining scenario matching in a process knowledge base based on the machining position time-series slice to obtain a historical control parameter sequence; fusing the machining position time-series slice and the high-frequency real-time sensor data stream to output a multimodal real-time operating condition feature tensor; performing incremental prediction of discharge parameters based on the multimodal real-time operating condition feature tensor and the historical control parameter sequence, and outputting an updated pulse parameter adjustment vector; the edge intelligent platform sending the updated pulse parameter adjustment vector to a PLC controller to drive the EDM machine's discharge module to perform smooth parameter tuning transition; and iteratively executing closed-loop prediction parameter tuning of the EDM machine's discharge module based on the dynamic sliding window update of the high-frequency real-time sensor data stream and the machining position time-series slice.

[0005] The second aspect of this application discloses a control system for an electrical discharge machining (EDM) machine based on industrial data processing. This system is used in the aforementioned EDM machine control method based on industrial data processing. The system includes: a working condition capture unit, used to synchronously capture working conditions using a multi-physics field sensor array during the movement of the EDM machine and output a high-frequency real-time sensor data stream; a sensor data receiving unit, used by an edge intelligent platform to receive the high-frequency real-time sensor data stream via a redundant dual-channel RS485 bus and synchronously retrieve machining position timing slices from the closed-loop position feedback module of the CNC system; and a machining scene matching unit, used to match isomorphic machining scenes in a process knowledge base based on the machining position timing slices. The system is configured to: obtain a historical control parameter sequence; a data fusion unit, which fuses the processing position time-series slices and high-frequency real-time sensor data streams to output a multimodal real-time operating condition feature tensor; an incremental prediction unit, which performs incremental prediction of discharge parameters based on the multimodal real-time operating condition feature tensor and the historical control parameter sequence to output an updated pulse parameter adjustment vector; a smooth parameter tuning transition unit, which sends the updated pulse parameter adjustment vector to the PLC controller via the edge intelligent platform to drive the machine discharge module to perform a smooth parameter tuning transition; and a closed-loop prediction parameter tuning unit, which iteratively executes the closed-loop prediction parameter tuning of the machine discharge module based on the dynamic sliding window update of the high-frequency real-time sensor data stream and the processing position time-series slices.

[0006] One or more technical solutions provided in this application have at least the following beneficial effects: By deploying multi-physics field sensor arrays on key parts of the EDM machine, dynamic working condition data during the machining process can be captured in real time. These high-frequency real-time data streams can reflect changes in the machining process, thus providing accurate data support for subsequent intelligent control and adjustment. The use of redundant dual-channel RS485 buses can enhance the reliability of data transmission. By transmitting data through dual channels simultaneously, data loss or transmission interruption caused by a single channel failure can be prevented. In addition, the edge intelligent platform receives high-frequency data streams in real time and synchronously retrieves the time-series slices of the machining position with the closed-loop position feedback module of the CNC system. This provides more accurate positioning information for subsequent analysis. By matching the machining position time-series slices with isomorphic machining scenarios in the process knowledge base, similar machining tasks and control parameters can be found based on historical data. This matching process can quickly obtain historical experience and optimized control parameter sequences, thus providing a more suitable starting point for the current machining task, improving machining efficiency, and avoiding trial-and-error adjustments from scratch. The system integrates processing position time-series slices with high-frequency sensor data streams to generate a multimodal real-time operating condition feature tensor, which comprehensively reflects the real-time status of the processing process from multiple perspectives. By combining the real-time operating condition feature tensor with historical control parameter sequences, an incremental prediction algorithm is used to calculate an accurate update pulse parameter adjustment vector. This not only reduces manual intervention but also dynamically optimizes control parameters according to changes in operating conditions, improving the stability and accuracy of the processing process. The update pulse parameter adjustment vector is sent to the PLC controller through an edge intelligent platform, driving the machine's discharge module to perform a smooth parameter adjustment transition. This process ensures that the adjustment of discharge parameters is smooth and continuous, avoiding processing instability or inaccuracy caused by sudden or rapid parameter changes. Through dynamic sliding window updates of real-time sensor data and processing position time-series slices, real-time iterative execution of closed-loop predictive parameter adjustment can be performed. This iterative process ensures that the parameter adjustment of each discharge process is optimized, enabling the processing process to maintain high efficiency and accuracy over a long period of time.

[0007] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0008] Figure 1 A schematic flowchart of an electrical discharge machining (EDM) control method based on industrial data processing provided in an embodiment of this application.

[0009] Figure 2 A schematic diagram of the control system structure for an electrical discharge machining (EDM) machine based on industrial data processing, provided in an embodiment of this application.

[0010] Explanation of reference numerals in the attached figures: 10 for working condition capture unit, 20 for sensor data receiving unit, 30 for processing scene matching unit, 40 for data fusion unit, 50 for incremental prediction unit, 60 for smooth parameter tuning and transition unit, and 70 for closed-loop prediction and parameter tuning unit. Detailed Implementation

[0011] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0012] Example 1, as Figure 1 As shown in the figure, this application provides a control method for an electrical discharge machining (EDM) machine based on industrial data processing. The method includes: During the movement of the electrical discharge machining (EDM) machine, a synchronous multi-physics field sensor array captures the working conditions and outputs a high-frequency real-time sensor data stream.

[0013] During the operation of an electrical discharge machining (EDM) machine, complex electrical discharge phenomena occur between the electrode and the workpiece, accompanied by different physical processes such as current, temperature, and vibration. In order to comprehensively capture these working conditions, a multi-physics field sensor array is used. The sensor composition includes a broadband Hall current sensor, a triaxial MEMS broadband accelerometer, and a PT100 platinum resistance temperature sensor. The data captured by these sensors is processed and output in the form of a high-frequency real-time data stream, providing real-time working condition information for subsequent processing stages.

[0014] The edge intelligent platform receives the high-frequency real-time sensing data stream through a redundant dual-channel RS485 bus and synchronously retrieves the machining position timing slice from the closed-loop position feedback module of the CNC system.

[0015] The edge intelligent platform receives and processes large amounts of real-time data. It performs tasks such as data preprocessing, analysis, and forwarding, and has computing power to perform complex real-time feature extraction and decision-making. The RS485 bus is a communication protocol used in industrial automation due to its strong anti-interference capabilities and long-distance transmission advantages. In this step, a redundant dual-channel RS485 bus ensures reliable data transmission; even if one channel fails, the other can continue transmitting data. High-frequency real-time sensor data streams are transmitted to the edge intelligent platform via the redundant dual-channel RS485 bus. The closed-loop position feedback module is part of the CNC system and is used to provide real-time feedback on the machine tool's position information. The edge intelligent platform communicates with this closed-loop position feedback module to obtain a time-series slice of the current machining position. This position data includes the machine tool's position changes at each moment, reflecting the workpiece's machining trajectory and key parameters during the machining operation.

[0016] Based on the time-series slices of the processing location, isomorphic processing scenarios are matched in the process knowledge base to obtain the historical control parameter sequence.

[0017] The process knowledge base is a database containing historical machining tasks. It stores parameters, operating procedures, and process settings for different machining scenarios. This information is derived from years of machining experience and data accumulation. Isomorphic machining scenario matching means comparing the machining location's time-series slices with historical data in the process knowledge base to find the most similar machining scenarios. Specifically, distance-based similarity calculations are used to determine which parameter combinations are most similar to those in historical scenarios at the current machining location. This matching method retrieves historical control parameter sequences from the process knowledge base. These sequences include multiple process parameters such as current, pulse frequency, working fluid temperature, and discharge time, helping to guide parameter tuning for the current machining task.

[0018] By fusing the time-series slices of the processing location and the high-frequency real-time sensor data stream, a multimodal real-time operating condition feature tensor is output.

[0019] By combining time-series slices of the machining position with high-frequency real-time sensor data streams, a comprehensive multimodal data stream is formed. This data not only includes the motion state of the mechanical equipment but also reflects environmental changes during operation. The fused data constitutes a multimodal real-time operating condition feature tensor, where each dimension represents a different type of data, comprehensively representing the characteristics of the machining machine under different operating conditions and providing high-precision operating condition information.

[0020] Based on the multimodal real-time operating condition feature tensor and historical control parameter sequence, the discharge parameter increment is predicted, and the updated pulse parameter adjustment vector is output.

[0021] By utilizing multimodal real-time operating condition feature tensors and historical control parameter sequences, new discharge parameter adjustments are predicted through machine learning or data-driven prediction models, such as time-series prediction models. The prediction result is an updated pulse parameter adjustment vector, which contains how to adjust the current discharge parameters, such as increasing or decreasing the pulse width, current, etc.

[0022] The edge intelligent platform sends the updated pulse parameter adjustment vector to the PLC controller, driving the machine discharge module to perform a smooth parameter adjustment transition.

[0023] As a central processing platform, the edge intelligence platform sends the predicted updated pulse parameter adjustment vector to the PLC (Programmable Logic Controller) of the machining machine. The PLC is a crucial controller in industrial automation, used to control and adjust the machining machine's actions in real time. The PLC controller changes the discharge parameters during the machining process by receiving adjustment commands from the edge intelligence platform. To avoid abrupt changes or instability caused by parameter adjustments, a smooth parameter adjustment transition mechanism is designed. This means that when adjusting pulse parameters, the PLC gradually adjusts the relevant discharge current, pulse frequency, etc., ensuring a smooth transition during the machining process and preventing excessive vibration, overheating, or other unstable factors from adversely affecting the workpiece. This smooth transition is achieved through techniques such as PID control and fuzzy control.

[0024] Based on the dynamic sliding window update of the high-frequency real-time sensing data stream and the processing position time slice, the closed-loop prediction parameter tuning of the processing machine discharge module is iteratively executed.

[0025] Sliding window technology is a method for processing time-series data. In this step, a dynamic sliding window is used to continuously acquire high-frequency real-time sensor data streams and processing location time-series slices over a recent period. The sliding window can be adjusted according to different needs and real-time data streams. For example, the sliding window length can be dynamically adjusted according to changes in operating conditions in order to more accurately capture real-time changing operating condition characteristics.

[0026] Closed-loop predictive parameter tuning is a closed-loop control process, meaning that the adjustment of the EDM machine's discharge parameters is not completed in one go, but is a continuous iterative process. The control model is adjusted based on real-time data feedback, and parameter prediction is performed again after each adjustment. Fine-tuning is then carried out based on the EDM machine's performance feedback. This forms an adaptive and continuously optimized process, which ensures precise control and optimization of the EDM machine during the machining process, thereby improving machining efficiency, reducing machining errors, and ensuring workpiece quality.

[0027] Furthermore, based on the time-series slices of the processing location, isomorphic processing scenarios are matched in the process knowledge base to obtain a historical control parameter sequence. The method includes: Based on the time-series slices of the processing position, a 3D processing trajectory point cloud is reconstructed to obtain a spatial pose topology map. Then, differential geometric invariant features are extracted to obtain a geometric topology feature vector. Based on the time-series slices of the processing position, motion state parameters are statistically analyzed, and the feed kinematic feature matrix is ​​output. Material attribute tags are retrieved according to the real-time workpiece code. The geometric topology feature vector, the feed kinematic feature matrix, and the material attribute tags are cascaded and fused to generate a scene fingerprint code. Using the scene fingerprint code as the index key, a high-dimensional feature space nearest neighbor search is performed in the process knowledge base to output the historical control parameter sequence.

[0028] Spatial modeling is performed based on time-series slices of the machining position to reconstruct a 3D trajectory point cloud of the machining machine. This point cloud represents the machine's position distribution at various moments during the machining process. Using this point cloud data, the path and trajectory of the machining machine in 3D space can be reconstructed. Through 3D point cloud reconstruction, a spatial pose topology map of the machining machine is obtained, representing its position and orientation in 3D space, as well as spatial changes during the machining process. Extracting invariant features using differential geometry methods refers to extracting features that remain unchanged under spatial transformations from the geometric shape of the machining trajectory. Specific geometric invariant features include: radius of curvature, representing the degree of curvature of the curve and reflecting changes in the curvature of the machining trajectory; aspect ratio, representing the ratio of the depth to the width of the machining area, used to describe the shape of the machining area; and tilt angle, representing the angle between the machining trajectory and the horizontal plane, reflecting changes in the machine's attitude.

[0029] Statistical analysis of the machining process is performed, primarily by calculating kinematic characteristics through time-series slices of the machining position. These kinematic characteristics provide information about how the machining machine moves, including feed rate, acceleration, and the rate of change of trajectory curvature. Feed rate refers to the speed at which the machining machine moves along the trajectory during machining, directly affecting machining accuracy and surface quality. Acceleration describes the acceleration and deceleration process of the machining machine on the trajectory; excessive acceleration may lead to vibration, thus affecting machining quality. Trajectory curvature refers to the rate of change of the curvature of the machining trajectory. The rate of change of the curvature of the machining trajectory reflects the complexity and dynamic changes of its motion path, especially when the machining path undergoes a sharp turn, the rate of change of trajectory curvature will increase.

[0030] In actual processing, workpieces are coded for easy identification and tracking. Each workpiece has different material properties, shape requirements, etc., and the workpiece code is used to identify this information. Based on the real-time workpiece code, the material attribute tags corresponding to that workpiece are retrieved from the database or material management system, including material type, physical properties of the material, processing requirements, etc. These material attributes have a significant impact on the selection of processing parameters.

[0031] The geometric topological feature vector, the feed kinematic feature matrix, and the material property labels are fused to generate a comprehensive feature vector, called scene fingerprint encoding. This process involves splicing different types of data to form a high-dimensional feature vector containing all key information. The generated scene fingerprint encoding is a high-level summary of the processing scene.

[0032] Using scene fingerprint codes as query keywords, a search is performed in the process knowledge base. Nearest neighbor search refers to finding the most matching historical sample in a high-dimensional feature space by calculating the similarity between scene fingerprint codes and historical data. By searching for similar scenes, the closest historical processing task is found, and the corresponding historical control parameter sequence is extracted. This historical control parameter sequence can be used as a reference for the current processing task.

[0033] Furthermore, the edge intelligent platform receives the high-frequency real-time sensor data stream via a redundant dual-channel RS485 bus and synchronously retrieves machining position timing slices from the closed-loop position feedback module of the CNC system. The method includes: The high-frequency real-time sensing data stream is parallel-loaded onto the redundant dual-channel RS485 bus and transmitted to the edge intelligent platform using a timestamp-aligned encapsulation protocol. The receiving end of the edge intelligent platform locates the main data source channel by calculating the CRC-32 checksum of the dual-channel data and then performs seamless switching in case of channel failure to receive the high-frequency real-time sensing data stream. A position sampling trigger pulse is sent to the closed-loop position feedback module of the CNC system according to the acquisition time window of the high-frequency real-time sensing data stream. When the closed-loop position feedback module latches the three-dimensional absolute coordinates of the grating ruler at the rising edge of the position sampling trigger pulse, the edge intelligent platform simultaneously freezes the multi-physics sensing data stream buffer. A machining position timing slice is generated based on the three-dimensional absolute coordinates of the grating ruler and the multi-physics sensing data stream.

[0034] High-frequency real-time sensor data streams are transmitted to a redundant dual-channel RS485 bus via parallel offloading. Redundancy effectively prevents data loss or communication interruptions caused by a single channel failure, ensuring stable and continuous data transmission. Since the high-frequency real-time sensor data streams originate from multiple sensors, these data may not be perfectly synchronized in time. To ensure time consistency, a timestamp is added to each data entry. Through timestamp alignment, real-time data from all sensors can be encapsulated and transmitted according to a unified time base. This approach ensures that the data stream maintains its chronological order during transmission, facilitating subsequent data processing and analysis.

[0035] CRC-32 (Cyclic Redundancy Check) is an error detection algorithm used to detect data corruption during transmission. After receiving data, the edge intelligent platform calculates the CRC checksum of the data packet and compares it with the checksum provided by the sender. If the checksums do not match, it indicates an error occurred during transmission. The edge intelligent platform receives data from two channels. By analyzing the CRC checksum results, it can determine which channel's transmission is valid, thus identifying the primary data source channel. During real-time data transmission, data continuity and stability must be ensured. If one transmission channel fails, such as due to signal loss or erroneous data, it seamlessly switches to the other functioning channel, ensuring no sensor data is lost and real-time control and monitoring remain unaffected.

[0036] High-frequency real-time sensor data streams are acquired in fixed time windows, such as a certain number of data points per second. This time window is determined based on the data stream's sampling frequency and the system's real-time requirements. According to the acquisition time window of the high-frequency real-time sensor data stream, a position sampling trigger pulse is sent to the closed-loop position feedback module of the CNC system. This pulse informs the CNC system that the machine tool's position needs to be sampled, i.e., the absolute position at the current moment is latched. This position sampling trigger pulse triggers a position sampling action, which can be achieved using a linear encoder or other positioning devices to obtain the machine tool's three-dimensional spatial position.

[0037] When the rising edge of the position sampling trigger pulse arrives, the closed-loop position feedback module in the CNC system latches the current position of the machining center, that is, records the three-dimensional absolute coordinates provided by its grating ruler. The grating ruler is a high-precision position measurement device that can provide accurate three-dimensional position data of the machining center, including the position and angle information of the X, Y, and Z axes. Synchronously with position sampling, the edge intelligent platform freezes the multiphysics sensor data stream buffer. Freezing the data stream ensures that all sensor data corresponds to the current machining center position during current position sampling. This synchronizes the correspondence between sensor data and machining position, providing accurate data for subsequent machining parameter prediction and optimization.

[0038] The obtained three-dimensional absolute coordinates of the grating ruler and the multi-physics field sensing data stream are combined. By synchronizing the position of the processing machine with the data of various sensors, the sensor status corresponding to each processing position is obtained. These data are organized into processing position time slices in chronological order. That is, the processing position and the corresponding sensor data in each time window form a time slice.

[0039] Furthermore, based on the multimodal real-time operating condition feature tensor and historical control parameter sequence, the method performs incremental prediction of discharge parameters and outputs an updated pulse parameter adjustment vector, comprising: Based on a preset time window, the multimodal real-time operating condition feature tensor and historical control parameter sequence are aligned and spatiotemporally sliced ​​to obtain a feature tensor quantum sequence and a control parameter subsequence. The feature tensor quantum sequence is flattened and compressed, and after being converted into a two-dimensional feature vector sequence, the two-dimensional feature vector sequence and the control parameter subsequence are concatenated and spliced ​​to construct a fused input vector sequence. The fused input vector sequence is loaded into a bidirectional LSTM time-series differential predictor to perform incremental prediction of discharge parameters and output the updated pulse parameter adjustment vector.

[0040] A preset time window is a specific time period during which data is considered a complete processing unit. The length of the preset time window is set according to the periodicity of the processing and the required response speed to ensure that key features related to the processing can be captured.

[0041] Spatiotemporal alignment refers to aligning multimodal real-time operating condition feature tensors and historical control parameter sequences along the time dimension. Since each type of data is collected at different points in time or time intervals, they need to be aligned using timestamps to ensure they represent the operating conditions at the same or similar times. The aligned data is divided into aligned spatiotemporal slices according to a preset time window, meaning that the operating condition features and control parameter data within each slice are within the same time period. The feature tensor quantum sequence refers to a fragment of feature tensor composed of all sensor data within a time window; the control parameter subsequence is the parameter subsequence extracted from historical control data within the corresponding time window.

[0042] Flattening compression is the process of transforming high-dimensional data into low-dimensional data. Here, the feature tensor quantum sequence contains multi-dimensional data. The purpose of flattening compression is to compress this multi-dimensional data into low-dimensional feature vectors. Flattening compression helps to simplify subsequent processing and improve computational efficiency, while maintaining important feature information. The flattened compressed data is transformed into a two-dimensional feature vector sequence, that is, the data of each time window is represented as a vector, and each time window corresponds to a feature vector, thus forming a feature matrix of a time series.

[0043] The two-dimensional feature vector sequence and the control parameter subsequence are concatenated and spliced ​​together. Concatenation and splicing refers to merging the two types of data into a long vector, called the fused input vector sequence. In this way, the model input will simultaneously contain operating condition data and historical control data, enabling the model to utilize both types of information at the same time.

[0044] A bidirectional LSTM is a special type of LSTM network that can learn not only the forward information of the current time step but also the contextual information of future time steps through backpropagation. This bidirectional structure allows LSTM to better capture the long-term and short-term dependencies in time series, making it particularly effective for time-series data. The fused input vector sequence is loaded into a bidirectional LSTM time-series differential predictor. The core objective of the predictor is to predict the incremental changes in discharge parameters based on the current multimodal operating conditions and historical control parameters. By analyzing the relationship between the operating conditions and control data, the predictor can predict the increment of control parameters at the next moment, such as the increase or decrease of pulse current. Differential prediction can be understood as predicting the increment of a parameter change, rather than directly predicting the absolute value. This allows the system to make fine-grained adjustments and more accurately control the processing. Finally, after prediction by the bidirectional LSTM time-series differential predictor, an updated pulse parameter adjustment vector is obtained. This vector contains adjustment instructions for each pulse parameter during the discharge process.

[0045] Furthermore, the method involves loading the fused input vector sequence into a bidirectional LSTM time-series differential predictor to perform incremental prediction of discharge parameters and outputting the updated pulse parameter adjustment vector. Retrieve the spatiotemporally aligned working condition feature tensor sequence and control parameter evolution sequence from the historical processing task log; construct a historical input vector sequence based on the working condition feature tensor sequence and control parameter evolution sequence; perform synchronous time window difference operation on the control parameter evolution sequence to generate a historical incremental truth sequence; use the historical input vector sequence as training data and the historical incremental truth sequence as supervision labels to perform spatiotemporal gradient backpropagation iterative parameter tuning of the bidirectional LSTM temporal differential predictor.

[0046] Historical processing task logs record key data from past processing processes, including changes in operating conditions and control parameters. When processing historical data, the operating conditions and control parameters are aligned along the time axis. The operating condition feature tensor sequence contains multimodal sensor data at each moment, forming a high-dimensional matrix that records the operating conditions at certain moments or time windows. The control parameter evolution sequence records the evolution trajectory of each control parameter during the discharge process at the corresponding moment or time window.

[0047] The operating condition feature tensor sequence and the control parameter evolution sequence are transformed into a historical input vector sequence. The specific operation is similar to the previous steps of constructing the fused input vector sequence, namely: the operating condition feature tensor sequence and the control parameter evolution sequence are aligned and spatiotemporally sliced ​​to ensure that the slices of sensor data and control parameters within the same time window are corresponding; the operating condition feature tensor sequence is flattened and compressed to convert high-dimensional data into low-dimensional feature vectors; the flattened and compressed operating condition feature vectors and control parameter vectors are concatenated together to form a historical input vector sequence for the model to learn more comprehensively.

[0048] Time window differential operation refers to the differential processing of the evolution sequence of control parameters to calculate the change of parameters between adjacent time points. This process can reveal the incremental characteristics of parameter changes. When performing differential operation, a synchronous time window is used, that is, the parameter difference within the same time window is calculated to ensure that the magnitude of the change is based on the same time reference point. In this way, it is possible to clarify how each control parameter evolves over time. The historical increment true value sequence generated by differential operation represents the incremental change of control parameters over time.

[0049] Historical input vector sequences are used as training data and fed into a bidirectional LSTM time-series differential predictor. This data is used to learn the complex relationship between operating conditions and control parameters, helping the model predict future discharge parameter increments. For each time point during training, the historical increment ground value sequence is used as a supervision label, i.e., the target value. The model's task is to predict results as close as possible to these historical increment ground value sequences through training. During training, spatiotemporal gradient backpropagation updates the gradients based on the error between the model's prediction and the historical increment ground value sequence. The goal of this process is to continuously optimize the model parameters so that the model can minimize errors during prediction, thereby improving prediction accuracy. Through backpropagation, the model's weights and biases are updated, enabling the model to better learn the complex relationships in the time series and effectively predict future discharge parameter increments.

[0050] Furthermore, the edge intelligent platform sends the updated pulse parameter adjustment vector to the PLC controller, driving the machine discharge module to perform a smooth parameter adjustment transition. The method includes: The edge intelligent platform serializes and encodes the updated pulse parameter adjustment vector into a control command frame; it sends the control command frame to the PLC controller via the industrial real-time Ethernet protocol; after parsing the command frame, the PLC controller drives the electrode vibration suppressor and discharge pulse generator of the machine discharge module to perform first-order smooth transition parameter tuning, and performs millisecond-level ultra-short-term adaptive pulse parameter within the preset process feasible domain.

[0051] To ensure effective data transmission between different systems, the edge intelligence platform serializes and encodes the updated pulse parameter adjustment vector into control command frames. This process essentially converts the data into a format suitable for transmission, such as a byte stream (e.g., JSON) or binary encoding. The control command frame contains the specific values ​​for the pulse parameter adjustments, as well as other control command information, such as the device ID and command identifier. This data needs to be parsed and executed by the receiving end.

[0052] The Industrial Real-Time Ethernet protocol is used for high-speed data transmission between devices. Here, the edge intelligent platform sends control command frames through the Industrial Real-Time Ethernet protocol to ensure that data can be transmitted to the PLC controller quickly and stably. Real-time transmission is crucial in this process because in industrial applications, especially in the process of electrical discharge machining, parameters need to be adjusted quickly and accurately to ensure processing quality and stability.

[0053] When the PLC controller receives a control command frame, it first parses it. This parsing process includes reading the data in the control frame and determining the pulse parameter adjustments that need to be performed. Based on the parsed instructions, the PLC controller controls two key components of the machining center's discharge module: the electrode vibration suppressor (electrode vibration affects the accuracy and stability of the discharge), and the discharge pulse generator (the PLC controller adjusts the parameters of the electrode vibration suppressor to suppress unnecessary vibration and improve machining accuracy). The discharge pulse generator is responsible for generating discharge pulses. The PLC controller adjusts the parameters of the pulse generator to ensure that parameters such as pulse current and discharge width reach the ideal machining state. First-order smooth transition parameter tuning means that when adjusting pulse parameters, the PLC controller does not immediately and drastically change the parameters, but rather adjusts them gradually through a smooth transition to avoid drastic fluctuations or instability in the machining process. This prevents a decrease in machining quality or equipment damage due to sudden parameter changes.

[0054] In practice, electrical discharge machining (EDM) requires parameter adjustments within a certain process-feasible domain. The process-feasible domain refers to the range of parameters that have been verified and can ensure machining quality and stable equipment operation. Millisecond-level ultra-short-term adaptive operation means that the discharge parameters can be adjusted in an extremely short time, usually at the millisecond level, and that these adjustments are based on real-time feedback. This allows the system to respond quickly to any external disturbances or internal changes, such as changes in workpiece material properties or machining position, while maintaining the stability and accuracy of the machining process.

[0055] Furthermore, the method also includes: After the real-time workpiece processing is completed, the closed-loop parameter tuning timing record is retrieved from the machine discharge module. The closed-loop parameter tuning timing record includes the response curve of the electrode vibration suppressor and the pulse parameter adjustment trajectory. Based on the closed-loop parameter tuning timing record, the machine offset fault trend is predicted, and the dynamic offset fault vector between axes is output.

[0056] After completing the real-time EDM machining task, the closed-loop parameter tuning timing record is retrieved from the EDM module. This data records the pulse parameters adjusted each time during the machining process, as well as the corresponding system response, for subsequent analysis and fault diagnosis. Among them, the suppression of electrode vibration is a critical parameter in the EDM process. Excessive vibration may lead to a decrease in machining accuracy. The electrode vibration suppressor response curve records the response state of the electrode vibration suppressor, revealing the trend of electrode vibration changes. The pulse parameter adjustment trajectory records the adjustment trajectory of the pulse parameters during the machining process. By analyzing this trajectory, it can be determined how the pulse parameters change over time during the machining process, and the flexibility and stability of the adjustment.

[0057] Machining machine misalignment faults refer to positional shifts that occur during machining due to factors such as wear of mechanical parts, uneven load, and loosening of mechanical structures. These shifts directly affect the machining accuracy and quality of the workpiece, and can even lead to machining failure. Analysis of the response curve of the electrode vibration suppressor and the pulse parameter adjustment trajectory detects abnormal misalignments during machining. For example, irregular fluctuations in the response curve of the electrode vibration suppressor indicate mechanical misalignment or instability; unexpected changes in the discharge parameter adjustment trajectory also indicate displacement or malfunction in certain parts of the machining machine. The resulting inter-axis dynamic misalignment fault vector describes the misalignment faults between different axes of the machining machine, reflecting the misalignment trend and magnitude in each axis.

[0058] Furthermore, the sensor configuration of the multi-physics sensing array includes a broadband Hall current sensor mounted on the electrode power busbar, a triaxial MEMS broadband accelerometer integrated into the end flange of the electrode clamping mechanism, and a PT100 platinum resistance temperature sensor embedded in the wall of the working fluid return channel.

[0059] Hall current sensors are used to measure current. They sense changes in the magnetic field generated by the flow of current through the Hall effect principle and convert them into voltage signals. The Hall current sensor is installed at the electrode power busbar to monitor real-time changes in current during electrical discharge machining. Its wide bandwidth means that the sensor can operate over a wide frequency range and can capture current fluctuations at the microsecond level.

[0060] MEMS broadband accelerometers are used to measure acceleration. Here, a triaxial accelerometer is used, which can monitor vibration acceleration in three directions, namely the X, Y, and Z axes. The accelerometer is installed on the end flange of the electrode clamping mechanism and can directly sense the motion and vibration of the electrode.

[0061] The PT100 platinum resistance temperature sensor is a type of temperature sensor that uses the characteristic of the resistance of platinum material changing with temperature to measure temperature. In the electrical discharge machining process, the working fluid not only cools the electrodes but also removes debris and heat generated during machining. The temperature change of the working fluid directly affects the machining effect and quality. Therefore, the PT100 platinum resistance temperature sensor embedded in the working fluid return channel monitors the temperature and temperature rise of the working fluid in real time, providing information on cooling and thermal management.

[0062] Furthermore, the high-frequency real-time sensing data stream includes the microsecond-level pulse current waveform sequence acquired and transmitted by the Hall current sensor, the multi-axis vibration acceleration spectrum stream acquired and transmitted by the triaxial MEMS broadband accelerometer, and the subsecond-level working fluid temperature rise gradient time series data acquired and transmitted by the PT100 platinum resistance temperature sensor.

[0063] Since electrical discharge machining (EDM) essentially utilizes pulsed current for discharge, the current changes extremely rapidly. Hall effect current sensors capture microsecond-level pulsed current waveform sequences, accurately recording current variations in each discharge cycle. Multiaxial vibration acceleration spectrum flow refers to the spectral data of vibration signals acquired through accelerometers, representing the vibration intensity at different frequencies. In EDM, the vibration frequency and intensity of the electrode significantly impact machining accuracy; excessive vibration can lead to machining errors. Therefore, acquiring multiaxial vibration acceleration spectrum flow helps monitor electrode vibration in real time and make timely adjustments to ensure machining accuracy. Working fluid temperature rise gradient refers to the rate of change of working fluid temperature over time. Excessive temperature can cause electrode overheating, damage to the machined surface, and even equipment malfunction. Therefore, monitoring temperature changes is crucial. Sub-second time-series data refers to real-time monitoring of temperature changes, accurate to less than one second, reflecting the rapid temperature rise of the working fluid.

[0064] Example 2 is based on the same inventive concept as the EDM machine control method based on industrial data processing in the previous examples, such as... Figure 2 As shown in the figure, this application embodiment provides an electrical discharge machining (EDM) control system based on industrial data processing, the system comprising: The working condition capture unit 10 is used to synchronously capture the working condition of the multi-physics field sensor array during the movement of the EDM machine and output a high-frequency real-time sensor data stream; the sensor data receiving unit 20 is used by the edge intelligent platform to receive the high-frequency real-time sensor data stream through a redundant dual-channel RS485 bus and synchronously retrieve the machining position time sequence slice from the closed-loop position feedback module of the CNC system; the machining scene matching unit 30 is used to perform isomorphic machining scene matching in the process knowledge base based on the machining position time sequence slice to obtain the historical control parameter sequence; the data fusion unit 40 is used to fuse the machining position time sequence slice and the high-frequency real-time sensor data stream. The system comprises: a high-frequency real-time sensor data stream, outputting a multimodal real-time operating condition feature tensor; an incremental prediction unit 50, used to perform incremental prediction of discharge parameters based on the multimodal real-time operating condition feature tensor and historical control parameter sequences, and outputting an updated pulse parameter adjustment vector; a smooth parameter tuning transition unit 60, used by the edge intelligent platform to send the updated pulse parameter adjustment vector to the PLC controller, driving the machining machine discharge module to perform smooth parameter tuning transition; and a closed-loop prediction parameter tuning unit 70, used to iteratively execute the closed-loop prediction parameter tuning of the machining machine discharge module based on the dynamic sliding window update of the high-frequency real-time sensor data stream and the machining position time sequence slice.

[0065] Furthermore, the processing scene matching unit 30 is used to perform the following operation steps: Based on the time-series slices of the processing position, a 3D processing trajectory point cloud is reconstructed to obtain a spatial pose topology map. Then, differential geometric invariant features are extracted to obtain a geometric topology feature vector. Based on the time-series slices of the processing position, motion state parameters are statistically analyzed, and the feed kinematic feature matrix is ​​output. Material attribute tags are retrieved according to the real-time workpiece code. The geometric topology feature vector, the feed kinematic feature matrix, and the material attribute tags are cascaded and fused to generate a scene fingerprint code. Using the scene fingerprint code as the index key, a high-dimensional feature space nearest neighbor search is performed in the process knowledge base to output the historical control parameter sequence.

[0066] Furthermore, the sensing data receiving unit 20 is used to perform the following operation steps: The high-frequency real-time sensing data stream is parallel-loaded onto the redundant dual-channel RS485 bus and transmitted to the edge intelligent platform using a timestamp-aligned encapsulation protocol. The receiving end of the edge intelligent platform locates the main data source channel by calculating the CRC-32 checksum of the dual-channel data and then performs seamless switching in case of channel failure to receive the high-frequency real-time sensing data stream. A position sampling trigger pulse is sent to the closed-loop position feedback module of the CNC system according to the acquisition time window of the high-frequency real-time sensing data stream. When the closed-loop position feedback module latches the three-dimensional absolute coordinates of the grating ruler at the rising edge of the position sampling trigger pulse, the edge intelligent platform simultaneously freezes the multi-physics sensing data stream buffer. A machining position timing slice is generated based on the three-dimensional absolute coordinates of the grating ruler and the multi-physics sensing data stream.

[0067] Furthermore, the incremental prediction unit 50 is used to perform the following operation steps: Based on a preset time window, the multimodal real-time operating condition feature tensor and historical control parameter sequence are aligned and spatiotemporally sliced ​​to obtain a feature tensor quantum sequence and a control parameter subsequence. The feature tensor quantum sequence is flattened and compressed, and after being converted into a two-dimensional feature vector sequence, the two-dimensional feature vector sequence and the control parameter subsequence are concatenated and spliced ​​to construct a fused input vector sequence. The fused input vector sequence is loaded into a bidirectional LSTM time-series differential predictor to perform incremental prediction of discharge parameters and output the updated pulse parameter adjustment vector.

[0068] Furthermore, the incremental prediction unit 50 is used to perform the following operation steps: Retrieve the spatiotemporally aligned working condition feature tensor sequence and control parameter evolution sequence from the historical processing task log; construct a historical input vector sequence based on the working condition feature tensor sequence and control parameter evolution sequence; perform synchronous time window difference operation on the control parameter evolution sequence to generate a historical incremental truth sequence; use the historical input vector sequence as training data and the historical incremental truth sequence as supervision labels to perform spatiotemporal gradient backpropagation iterative parameter tuning of the bidirectional LSTM temporal differential predictor.

[0069] Furthermore, the smooth parameter tuning transition unit 60 is used to perform the following operation steps: The edge intelligent platform serializes and encodes the updated pulse parameter adjustment vector into a control command frame; it sends the control command frame to the PLC controller via the industrial real-time Ethernet protocol; after parsing the command frame, the PLC controller drives the electrode vibration suppressor and discharge pulse generator of the machine discharge module to perform first-order smooth transition parameter tuning, and performs millisecond-level ultra-short-term adaptive pulse parameter within the preset process feasible domain.

[0070] Furthermore, the closed-loop prediction and parameter tuning unit 70 is used to perform the following operation steps: After the real-time workpiece processing is completed, the closed-loop parameter tuning timing record is retrieved from the machine discharge module. The closed-loop parameter tuning timing record includes the response curve of the electrode vibration suppressor and the pulse parameter adjustment trajectory. Based on the closed-loop parameter tuning timing record, the machine offset fault trend is predicted, and the dynamic offset fault vector between axes is output.

[0071] Furthermore, the sensor configuration of the multi-physics sensing array includes a broadband Hall current sensor mounted on the electrode power busbar, a triaxial MEMS broadband accelerometer integrated into the end flange of the electrode clamping mechanism, and a PT100 platinum resistance temperature sensor embedded in the wall of the working fluid return channel.

[0072] Furthermore, the high-frequency real-time sensing data stream includes the microsecond-level pulse current waveform sequence acquired and transmitted by the Hall current sensor, the multi-axis vibration acceleration spectrum stream acquired and transmitted by the triaxial MEMS broadband accelerometer, and the subsecond-level working fluid temperature rise gradient time series data acquired and transmitted by the PT100 platinum resistance temperature sensor.

[0073] Through the foregoing detailed description of the control method for an electrical discharge machining (EDM) machine based on industrial data processing, those skilled in the art can clearly understand the EDM machine control system based on industrial data processing in this embodiment. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant parts can be referred to in the method section.

[0074] 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 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 control method for electrical discharge machining (EDM) machines based on industrial data processing, characterized in that, The method includes: During the movement of the electrical discharge machining machine, a synchronous multi-physics field sensor array captures the working conditions and outputs a high-frequency real-time sensor data stream. The edge intelligent platform receives the high-frequency real-time sensor data stream through a redundant dual-channel RS485 bus and synchronously retrieves the machining position timing slice from the closed-loop position feedback module of the CNC system. Based on the time-series slices of the processing location, isomorphic processing scenarios are matched in the process knowledge base to obtain the historical control parameter sequence; By fusing the processing location time-series slices and high-frequency real-time sensing data streams, a multimodal real-time operating condition feature tensor is output. Based on the multimodal real-time operating condition feature tensor and historical control parameter sequence, the discharge parameter increment is predicted, and the updated pulse parameter adjustment vector is output. The edge intelligent platform sends the updated pulse parameter adjustment vector to the PLC controller, driving the machine discharge module to perform a smooth parameter adjustment transition. Based on the dynamic sliding window update of the high-frequency real-time sensing data stream and the processing position time slice, the closed-loop prediction parameter tuning of the processing machine discharge module is iteratively executed.

2. The electrical discharge machining control method based on industrial data processing as described in claim 1, characterized in that, Based on the time-series slices of the processing positions, isomorphic processing scenarios are matched in the process knowledge base to obtain a sequence of historical control parameters. The method includes: Based on the time-series slices of the processing position, a three-dimensional processing trajectory point cloud is reconstructed to obtain a spatial pose topology map. Then, differential geometric invariant features are extracted to obtain geometric topology feature vectors. Based on the time slice of the processing position, the motion state parameters are statistically analyzed, and the kinematic feature matrix is ​​output. Retrieve material attribute tags based on real-time workpiece code; The geometric topological feature vector, the feed kinematic feature matrix, and the material property labels are cascaded and fused to generate a scene fingerprint code; Using the scenario fingerprint code as the index key, a high-dimensional feature space nearest neighbor search is performed in the process knowledge base to output the historical control parameter sequence.

3. The electrical discharge machining control method based on industrial data processing as described in claim 1, characterized in that, The edge intelligent platform receives the high-frequency real-time sensor data stream via a redundant dual-channel RS485 bus and synchronously retrieves machining position timing slices from the closed-loop position feedback module of the CNC system. The method includes: The high-frequency real-time sensing data stream is distributed in parallel to the redundant dual-channel RS485 bus and transmitted to the edge intelligent platform using a timestamp-aligned encapsulation protocol. The receiving end of the edge intelligent platform locates the main data source channel by calculating the CRC-32 checksum of the dual-channel data, and then performs seamless switching for channel failure in order to receive the high-frequency real-time sensing data stream. According to the acquisition time window of the high-frequency real-time sensing data stream, a position sampling trigger pulse is sent to the closed-loop position feedback module of the CNC system; When the closed-loop position feedback module latches the three-dimensional absolute coordinates of the grating ruler at the rising edge of the position sampling trigger pulse, the edge intelligent platform simultaneously freezes the multi-physics field sensing data stream buffer. Based on the three-dimensional absolute coordinates of the grating ruler and the multiphysics sensing data stream, a processing position time-series slice is generated.

4. The electrical discharge machining control method based on industrial data processing as described in claim 1, characterized in that, Based on the multimodal real-time operating condition feature tensor and historical control parameter sequence, the method performs incremental prediction of discharge parameters and outputs an updated pulse parameter adjustment vector, comprising: Based on a preset time window, the spatiotemporal slices of the multimodal real-time operating condition feature tensor and the historical control parameter sequence are aligned and extracted to obtain the feature tensor quantum sequence and the control parameter subsequence. The feature quantum sequence is flattened and compressed to output a two-dimensional feature vector sequence. The two-dimensional feature vector sequence and the control parameter subsequence are concatenated and spliced ​​to construct a fused input vector sequence. The fused input vector sequence is loaded into a bidirectional LSTM time-series differential predictor to perform incremental prediction of discharge parameters and output the updated pulse parameter adjustment vector.

5. The electrical discharge machining control method based on industrial data processing as described in claim 4, characterized in that, The method involves loading the fused input vector sequence into a bidirectional LSTM time-series differential predictor to perform incremental prediction of discharge parameters and outputting the updated pulse parameter adjustment vector. Retrieve the spatiotemporally aligned working condition feature tensor sequence and control parameter evolution sequence from historical processing task logs; A historical input vector sequence is constructed based on the aforementioned operating condition feature tensor sequence and control parameter evolution sequence; Perform synchronous time window difference operation on the evolution sequence of the control parameters to generate a historical incremental true value sequence; Using the historical input vector sequence as training data and the historical incremental true value sequence as supervision labels, the spatiotemporal gradient backpropagation iterative parameter tuning of the bidirectional LSTM temporal differential predictor is performed.

6. The electrical discharge machining control method based on industrial data processing as described in claim 1, characterized in that, The edge intelligent platform sends the updated pulse parameter adjustment vector to the PLC controller, driving the machine discharge module to perform a smooth parameter adjustment transition. The method includes: The edge intelligent platform serializes and encodes the updated pulse parameter adjustment vector into a control command frame; The control command frame is sent to the PLC controller via the industrial real-time Ethernet protocol. After parsing the instruction frame, the PLC controller drives the electrode vibration suppressor and discharge pulse generator of the machine discharge module to perform first-order smooth transition parameter tuning, and performs millisecond-level ultra-short-term adaptive pulse parameters within the preset process feasible domain.

7. The electrical discharge machining control method based on industrial data processing as described in claim 1, characterized in that, The method further includes: After the real-time workpiece processing is completed, the closed-loop parameter tuning timing record is retrieved from the discharge module of the processing machine. The closed-loop parameter tuning timing record includes the response curve of the electrode vibration suppressor and the pulse parameter adjustment trajectory. Based on the closed-loop parameter tuning timing record, the machine tool offset fault trend is predicted, and the dynamic offset fault vector between axes is output.

8. The electrical discharge machining control method based on industrial data processing as described in claim 1, characterized in that, The sensor configuration of the multiphysics field sensing array includes a broadband Hall current sensor mounted on the electrode power busbar, a triaxial MEMS broadband accelerometer integrated into the end flange of the electrode clamping mechanism, and a PT100 platinum resistance temperature sensor embedded in the wall of the working fluid return channel.

9. The electrical discharge machining control method based on industrial data processing as described in claim 8, characterized in that, The high-frequency real-time sensing data stream includes the microsecond-level pulse current waveform sequence acquired and transmitted by the Hall current sensor, the multi-axis vibration acceleration spectrum stream acquired and transmitted by the triaxial MEMS broadband accelerometer, and the subsecond-level working fluid temperature rise gradient time series data acquired and transmitted by the PT100 platinum resistance temperature sensor.

10. An electrical discharge machining (EDM) control system based on industrial data processing, characterized in that, For implementing the electrical discharge machining control method based on industrial data processing according to any one of claims 1-9, the system comprises: The working condition capture unit is used to synchronously capture the working condition of the multi-physics field sensor array during the movement of the electrical discharge machining machine and output a high-frequency real-time sensor data stream. The sensor data receiving unit is used by the edge intelligent platform to receive the high-frequency real-time sensor data stream through a redundant dual-channel RS485 bus, and synchronously retrieve the machining position timing slice from the closed-loop position feedback module of the CNC system. The processing scene matching unit is used to perform isomorphic processing scene matching in the process knowledge base based on the processing location time slice to obtain the historical control parameter sequence. The data fusion unit is used to fuse the processing location time slices and high-frequency real-time sensor data streams, and output a multimodal real-time operating condition feature tensor. The incremental prediction unit is used to perform incremental prediction of discharge parameters based on the multimodal real-time operating condition feature tensor and the historical control parameter sequence, and output an updated pulse parameter adjustment vector. A smooth parameter adjustment transition unit is used by the edge intelligent platform to send the updated pulse parameter adjustment vector to the PLC controller, driving the machine discharge module to perform a smooth parameter adjustment transition. The closed-loop prediction parameter tuning unit is used to iteratively execute the closed-loop prediction parameter tuning of the machining machine discharge module based on the dynamic sliding window update of the high-frequency real-time sensing data stream and the processing position time slice.

Images