Slow wire underarm automatic anti-collision device and method

Abstract

The application discloses a slow wire lower arm automatic anti-collision device and method, relates to the field of slow wire processing, and comprises a slow wire system module, which is used for issuing driving instructions and monitoring whether the real-time current of each feeding motion shaft exceeds the saved standard current threshold value and whether the duration thereof exceeds the saved standard time threshold value in the process of using a machine tool; is also used for executing an anti-collision self-learning program; a servo driving module is used for receiving the driving instructions of the slow wire system module and driving the linear motor module to move, while feeding back the current waveform curve of the linear motor of each feeding motion shaft; the linear motor module of each feeding motion shaft is used for moving each feeding motion shaft and generating a motor current signal; and a shutdown protection module is used for triggering shutdown when the real-time current of any feeding motion shaft exceeds the standard current threshold value and the duration thereof exceeds the standard time threshold value. The application has the effect of monitoring the motor current in real time and performing anti-collision prediction.

Description

Technical Field

[0001] This application relates to the field of wire EDM, and in particular to an automatic anti-collision device and method for the lower arm of a wire EDM machine. Background Technology

[0002] In the field of wire EDM machine tools, there are significant technical problems in the existing technology that urgently need to be solved. During wire threading or workpiece clamping operations, operators frequently encounter water jacket collisions and damage due to the difficulty in precisely controlling the distance between the water jacket and the workpiece. This makes the water jacket a frequently replaced consumable part, increasing spare parts costs and wasting considerable time on maintenance, severely impacting production efficiency. Furthermore, during machine operation, operator carelessness when moving the feed axes can easily cause the lower arm to collide with the workpiece. Such collisions have serious consequences, ranging from minor damage to the water jacket to severe damage to the lower arm structure, leading to high repair costs and extended equipment downtime, further exacerbating production efficiency losses. Currently, wire EDM machine tools generally lack effective collision avoidance mechanisms and cannot monitor and prevent these accidental collisions in real time. Therefore, an automated solution is urgently needed to predict and avoid collision risks, reduce maintenance costs, and improve equipment reliability. Summary of the Invention

[0003] To address the lack of real-time monitoring and collision avoidance prediction in traditional devices, this application provides an automatic anti-collision device and method for the lower arm of a slow wire EDM machine.

[0004] In the first aspect, the automatic anti-collision device for the lower arm of the slow wire EDM provided in this application adopts the following technical solution: including a slow wire EDM system module, a servo drive module, a linear motor module containing multiple feed motion axes, and a shutdown protection module; The slow wire EDM system module is located in the control core of the machine tool. It is used to issue drive commands and monitor in real time whether the real-time current of each feed axis exceeds the saved standard current threshold and whether its duration exceeds the saved standard time threshold during the use of the machine tool. It is also used to execute an anti-collision self-learning program, which includes an adaptive learning mechanism to dynamically adjust the standard current threshold and the standard time threshold according to the machine tool's historical current data and real-time status. The servo drive module is connected to the slow wire EDM system module and is used to receive the drive command from the slow wire EDM system module and drive the linear motor module to move, while feeding back the current waveform curve of the linear motor of each feed motion axis. The linear motor module, connected to the servo drive module, includes linear motors for each feed motion axis, used to realize the movement of each feed motion axis, generate motor current signals and feed them back through the servo drive module; The shutdown protection module is connected to the slow wire EDM system module and is used to trigger a shutdown and output an error message when the real-time current of any feed axis exceeds the standard current threshold and its duration exceeds the standard time threshold. During the execution of the anti-collision self-learning program, the slow wire EDM system module drives the servo drive module to move each feed motion axis. The peak current and rated current of each feed motion axis during normal axis movement are read through the current waveform curve fed back by the servo drive module. The standard current threshold and the standard time threshold for each feed motion axis to stop due to a collision are calculated and saved.

[0005] By adopting the above technical solution, the automatic anti-collision device for the lower arm of the wire EDM machine integrates the wire EDM system module, servo drive module, linear motor module, and shutdown protection module, realizing automated anti-collision monitoring and adaptive protection of the machine tool's feed motion axes. This solution monitors the current of each feed motion axis in real time through the wire EDM system module and executes an anti-collision self-learning program. Combined with the current waveform curve feedback from the servo drive module, it dynamically adjusts the standard current threshold and standard time threshold, ensuring the accuracy and adaptability of intelligent anti-collision judgment. The shutdown protection module automatically triggers shutdown and reports an error when it detects that the real-time current exceeds the standard current threshold and the duration exceeds the limit, effectively preventing collision damage to the wire EDM machine tool and reducing unplanned downtime and maintenance costs. The adaptive learning mechanism optimizes threshold settings based on historical current data and the real-time status of the machine tool, avoiding the tediousness and delay of manual threshold calibration, improving the intelligence level and response speed of the device, thereby significantly improving the operational safety, ease of operation, and overall reliability of the wire EDM machine tool, and supporting efficient preventive maintenance.

[0006] Optionally, the wire EDM system module includes a control command unit, a real-time monitoring unit, and an adaptive learning unit; The control instruction unit is used to generate drive instructions; The real-time monitoring unit is used to monitor the real-time current of each feed motion axis and perform current threshold comparison. The adaptive learning unit is used to run an embedded machine learning algorithm to analyze the machine tool's historical current data and real-time status to dynamically adjust and optimize the standard current threshold and the standard time threshold.

[0007] By adopting the above technical solution, the automatic anti-collision device for the lower arm of a wire EDM machine achieves in-depth optimization and synergistic effect of anti-collision protection function by refining the wire EDM system module into a control command unit, a real-time monitoring unit, and an adaptive learning unit. The control command unit is responsible for driving command generation, ensuring the accuracy and timeliness of motion control signals; the real-time monitoring unit independently performs continuous acquisition and threshold comparison of the current of each feed axis, significantly shortening the anomaly detection response cycle; the adaptive learning unit, based on embedded machine learning algorithms, autonomously iterates and optimizes the standard current threshold and standard time threshold through feature mining of historical current data of the machine tool and dynamic analysis of real-time working conditions, effectively solving the problem of false triggering or missed detection caused by the fluctuation of machining conditions due to traditional fixed thresholds. While ensuring the real-time performance of intelligent anti-collision judgment, it greatly improves the adaptability and accuracy of threshold setting, reduces the frequency of manual calibration, strengthens the device's generalization ability to different machining scenarios, provides a higher level of intelligent anti-collision protection for wire EDM machines, and further reduces equipment collision risks and maintenance costs.

[0008] Optionally, the servo drive module includes an instruction parsing unit, a current feedback unit, and a noise filtering unit; The instruction parsing unit is used to convert the drive instructions of the slow wire EDM system module into motor control signals; The current feedback unit is used to collect the current waveform curve of the linear motor module in real time and transmit it to the slow wire EDM system module. The noise filtering unit is used to apply digital filtering technology to eliminate the influence of environmental electromagnetic interference on the current waveform curve.

[0009] By adopting the above technical solution, the servo drive module of the automatic anti-collision device for the lower arm of the wire EDM machine is refined into a command parsing unit, a current feedback unit, and a noise filtering unit, significantly enhancing motion control accuracy and current signal reliability. The command parsing unit achieves efficient conversion of drive commands into motor control signals, ensuring real-time motion response and trajectory accuracy. The current feedback unit continuously captures the original current waveform curve of the linear motor module, providing millisecond-level dynamic data streams for the wire EDM system module. The noise filtering unit uses digital filtering technology to actively suppress environmental electromagnetic interference from contaminating the current waveform, eliminating the risk of false triggering. It ensures the integrity and authenticity of current feedback data under complex working conditions, enabling real-time and accurate monitoring and identification of true collision current characteristics, avoiding the problems of false shutdowns or missed alarms caused by signal distortion in traditional devices. This solution significantly improves the signal-to-noise ratio of the anti-collision criterion and the system robustness, extends the lifespan of core components, and provides underlying data support for the processing of the wire EDM machine.

[0010] Optionally, the linear motor module includes a motor drive unit, a position sensing unit, and a current generation unit; The motor drive unit is used to drive the feed motion axis to move according to the motor control signal of the servo drive module; The position sensing unit is used to detect the actual displacement of the feed motion axis and generate a position feedback signal; The current generation unit is used to convert the position feedback signal into a motor current signal and feed it back to the servo drive module.

[0011] By adopting the above technical solution, the linear motor module of the automatic anti-collision device for the lower arm of the wire EDM machine is refined into a motor drive unit, a position sensing unit, and a current generation unit, realizing a fully closed-loop collaborative optimization of motion control and state feedback. The motor drive unit accurately responds to the motor control signals of the servo drive module, ensuring the transient synchronization and trajectory stability of the feed motion axis displacement; the position sensing unit captures the actual displacement of the feed motion axis in real time and generates a position feedback signal, providing a data basis for motion error compensation; the current generation unit innovatively converts the position feedback signal into an equivalent motor current signal, building a seamless data link with the servo drive module. The linkage of the three units constructs a two-way mapping mechanism between displacement and current, enabling the wire EDM system module not only to monitor the current threshold but also to infer displacement anomalies through the current signal, identifying potential collision risks in advance. This solution eliminates the hysteresis defect of traditional single current monitoring, while reducing the cost of redundant sensor configuration, providing millisecond-level collision warnings for dynamically changing machining scenarios, and significantly improving the safety protection level and machining reliability of the wire EDM machine tool.

[0012] Optionally, the shutdown protection module includes a threshold judgment unit, a shutdown triggering unit, and an error output unit; The threshold judgment unit is used to receive the monitoring results of the slow wire EDM system module and confirm the current exceeding the standard event; The shutdown trigger unit is used to immediately cut off the power supply of the linear motor module and lock the feed motion axis corresponding to the current over-limit event after confirming the current over-limit event. The error output unit is used to generate detailed error information for the feed motion axis corresponding to the current over-limit event and display it through the human-machine interface.

[0013] By adopting the above technical solution, the shutdown protection module of the automatic anti-collision device for the lower arm of the wire EDM machine is refined into a hierarchical response mechanism consisting of a threshold judgment unit, a shutdown trigger unit, and an error output unit, thus constructing a collision protection system. The threshold judgment unit verifies the monitoring results of the wire EDM system module and accurately identifies current overrun events, eliminating the risk of misjudgment. After event confirmation, the shutdown trigger unit simultaneously cuts off the power supply to the linear motor module and physically locks the faulty feed axis, achieving instantaneous blocking of collision energy and mechanical hard protection. The error output unit automatically generates structured error information containing faulty axis location information, current overrun parameters, and a timestamp, which is presented intuitively through the human-machine interface. The serial execution of the three units forms a closed-loop process of event diagnosis, emergency braking, and fault tracing, transforming the traditional passive maintenance after a collision into active interception, significantly reducing the risk of mechanical damage to the equipment and spindle deformation. The function of accurately locating the faulty axis significantly shortens the mean time of repair, avoids production capacity loss caused by total machine downtime, and at the same time, the visualized error information provides operators with diagnostic basis, improving maintenance efficiency and production continuity.

[0014] Optionally, the adaptive learning unit also integrates a dynamic learning rate adjustment mechanism; the dynamic learning rate adjustment mechanism adjusts the learning rate of the embedded machine learning algorithm in real time based on the differences in the sensitivity of the processing material type and processing stage to current fluctuations. Specifically, when the hardness grade of the processed material increases or the processing stage changes, the learning rate automatically increases to accelerate the updating of the standard current threshold and the standard time threshold; when the hardness grade of the processed material decreases or the processing stage is in a stable cutting state, the learning rate automatically decreases to suppress noise interference.

[0015] By adopting the above technical solution, the adaptive learning unit of the automatic anti-collision device for the lower arm of wire EDM significantly enhances the adaptability and anti-interference capability of the threshold optimization function through the integration of a dynamic adjustment mechanism for the learning rate. This solution is based on the differentiated impact of changes in the hardness level of the processed material and the transition between processing stages on current fluctuation characteristics. It dynamically adjusts the learning rate of the embedded machine learning algorithm in real time. When the hardness of the processed material increases or the processing stage changes, the learning rate is automatically increased, accelerating the iterative update of the standard current threshold and standard time threshold, ensuring that the device quickly captures new current characteristic patterns when operating conditions change abruptly. When the material hardness decreases or the processing enters a stable cutting stage, the learning rate is actively reduced, effectively suppressing the interference of environmental noise and random fluctuations on threshold optimization. This design overcomes the response hysteresis defect of traditional fixed learning rates, significantly reducing the false trigger rate while ensuring adaptive accuracy. It particularly solves the problem of false alarms caused by transient current peaks in the roughing stage of high-hardness materials, and the risk of missed detection caused by noise accumulation in the stable finishing stage. By establishing an intelligent mapping relationship between material, operating condition, and rate, the optimal dynamic balance of threshold settings is maintained in complex and ever-changing processing scenarios.

[0016] Optionally, the adaptive learning unit adopts a reinforcement learning framework based on meta-learning, which includes a dynamic feature extraction layer and a policy transfer layer. The dynamic feature extraction layer uses a multi-scale convolutional neural network to simultaneously process the trend features of the machine tool's historical current data and the transient fluctuation features of the machine tool's real-time state, generating a current pattern vector that integrates spatiotemporal characteristics. The strategy migration layer constructs a digital twin of the machine tool operation based on the current mode vector, and pre-simulates the collision risk evolution path under different threshold combinations in the digital twin to select the reference threshold combination that minimizes the sum of false alarm rate and false negative rate. The adaptive learning unit initiates a reverse verification mechanism after each update of the reference threshold combination. It substitutes the current reference threshold combination into historical error event data for virtual replay. Only when the matching degree between the virtual replay result and the actual error event exceeds the preset verification standard will the reference threshold combination be applied to dynamic adjustment and optimization.

[0017] By adopting the above technical solution, the adaptive learning unit of the automatic anti-collision device for the lower arm of the wire EDM machine achieves a closed loop of predictive protection upgrade and reliability verification by introducing a reinforcement learning framework based on meta-learning. The dynamic feature extraction layer utilizes a multi-scale convolutional neural network to simultaneously extract the long-term trend features of historical current data of the machine tool and the transient fluctuation features of the real-time state, generating a current pattern vector that integrates spatiotemporal dimensions, breaking through the limitations of traditional single-dimensional data analysis. The strategy transfer layer constructs a high-fidelity digital twin based on this current pattern vector, and pre-enacts the collision risk evolution path under different threshold combinations in a virtual environment. Through intelligent game theory algorithms, it selects the optimal reference threshold combination with the minimum sum of false alarm rate and false negative rate, significantly improving the scientificity and foresight of threshold decision-making. The reverse verification mechanism is forcibly activated after each threshold update, substituting candidate thresholds into historical error events for virtual reenactment. The new reference threshold combination is only applied when the reenactment result matches the actual event, fundamentally avoiding the risk of threshold failure due to model overfitting or data bias. This solution is the first to achieve full-chain intelligent optimization in the field of collision avoidance, including feature fusion, digital pre-simulation, and reverse verification. It enables the device to have generalized adaptability to unknown working conditions and anti-interference resilience, significantly reducing the probability of collisions in complex machining and building a full life-cycle protection system with self-evolution capabilities for wire EDM machines.

[0018] Optionally, an environmental adaptation module is also included, connected to the linear motor module and the servo drive module, for real-time acquisition of ambient temperature data, and correction of the current waveform curve through a temperature compensation algorithm. At the same time, an unsteady-state heat transfer compensator is embedded to construct a differential geometric mapping relationship between the machine tool thermal deformation field and the current temperature drift. By solving the divergence term in the heat conduction equation in real time, the current baseline drift caused by temperature is separated. The temperature compensation algorithm adopts a Lyapunov exponential correction mechanism to dynamically compress the current monitoring lag window caused by sudden changes in ambient temperature.

[0019] By adopting the above technical solutions, the environmental adaptation module of the automatic anti-collision device for the lower arm of the wire EDM machine achieves a significant leap in the environmental anti-interference capability of current monitoring by integrating temperature data acquisition, temperature compensation algorithms, and an unsteady-state heat transfer compensator. This module acquires ambient temperature data in real time and applies a temperature compensation algorithm to correct the current waveform curve, eliminating signal distortion caused by temperature drift. The unsteady-state heat transfer compensator constructs a differential geometric mapping relationship between the machine tool's thermal deformation field and the current temperature drift. By solving the divergence term of the heat conduction equation, it accurately separates the current baseline drift caused by temperature, effectively solving the problem of false triggering or missed detection caused by temperature fluctuations in traditional devices. The Lyapunov exponential correction mechanism dynamically compresses the current monitoring lag window under sudden temperature changes, ensuring real-time responsiveness and avoiding delays in collision protection. This solution significantly improves the stability and accuracy of collision judgment in variable temperature environments, significantly reduces the rate of false shutdowns in high-precision machining, reduces thermal deformation damage and maintenance frequency, and enhances the device's adaptability to complex working conditions, providing reliable all-weather protection for wire EDM machines, thereby optimizing production continuity and the service life of the wire EDM machine.

[0020] Optionally, the environmental adaptation module further includes a load adaptive compensation unit, which is used to monitor the load change of the linear motor module in real time and dynamically adjust the parameters of the temperature compensation algorithm through a deep deterministic strategy gradient reinforcement learning algorithm to compensate for the additional thermal deformation caused by the load change; the output of the load adaptive compensation unit is used to update the differential geometric mapping relationship.

[0021] By adopting the above technical solution, the environmental adaptation module of the automatic anti-collision device for the lower arm of the wire EDM machine achieves a technological leap from single-dimensional temperature drift correction to dual-variable co-optimization of load and temperature through the addition of a load adaptive compensation unit. This unit monitors the dynamic load changes of the linear motor module in real time and uses a deep deterministic gradient reinforcement learning algorithm to iteratively adjust the key parameters of the temperature compensation algorithm online, accurately decoupling the additional thermal deformation interference caused by load fluctuations. Its output synchronously updates the topological structure of the differential geometric mapping relationship, ensuring that the solution results of the divergence term of the heat conduction equation synchronously reflect the coupling effect of load and temperature. This solution overcomes the theoretical limitation of traditional environmental compensation devices that ignore the influence of mechanical load on the thermal field, and solves the problem of distorted calculation of current baseline drift caused by heat accumulation under high-load cutting conditions. The dynamic parameter tuning mechanism of deep reinforcement learning compresses the temperature compensation lag window to the millisecond level, significantly improving the stability of the anti-collision threshold in variable load machining, reducing the false alarm rate and equipment thermal damage risk under complex working conditions, and providing all-element adaptive protection for wire EDM machines.

[0022] In a second aspect, the present invention provides an automatic anti-collision method for a slow wire EDM lower arm, applied to an automatic anti-collision device for a slow wire EDM lower arm provided in the first aspect. The method includes the following steps: S1. The anti-collision self-learning program is executed through the slow wire EDM system module, which drives the servo drive module to move the linear motor module of each feed motion axis, and reads the peak current and rated current of each feed motion axis during normal axis movement through the current data curve fed back by the servo drive module. S2. Based on the peak current and the rated current, calculate and save the standard current threshold and standard time threshold for each feed motion axis. S3. During the use of the machine tool, the real-time current of each feed axis is monitored in real time through the slow wire EDM system module; S4. When the real-time current of any feed axis exceeds the standard current threshold and its duration exceeds the standard time threshold, the shutdown protection module triggers a shutdown and outputs an error message.

[0023] The technical effects of the method of the present invention are the same as those of the first aspect, and will not be repeated here.

[0024] In summary, this application includes at least one of the following beneficial technical effects: 1. By employing an adaptive learning mechanism based on current waveform curves, the standard current threshold and standard time threshold are dynamically optimized through a collision avoidance self-learning program executed by the slow wire EDM system module, overcoming the response hysteresis defect of traditional fixed thresholds. Combined with embedded machine learning algorithms, real-time analysis of historical machine tool current data and operating conditions is performed. An innovative dynamic adjustment mechanism for the learning rate and a meta-learning reinforcement learning framework are introduced to automatically adjust the learning intensity based on material hardness and processing stage. The risk evolution path of threshold combinations is pre-enacted in a digital twin, and reliability is ensured through a reverse verification mechanism. This effectively solves the problems of low efficiency and poor threshold adaptability caused by manual calibration, leading to false triggering or missed detection. It significantly improves the accuracy and response speed of collision recognition, reducing the damage risk and maintenance costs of slow wire EDM machine tools. 2. By constructing a full-link collaborative mechanism of monitoring-drive-protection, electromagnetic interference is eliminated through the noise filtering unit of the servo drive module, displacement-current bidirectional mapping is achieved through the position sensing unit and current generation unit of the linear motor module, and the hierarchical response logic of the shutdown protection module forms a millisecond-level closed-loop protection system. This ensures the authenticity and integrity of the current signal under complex operating conditions, enabling the real-time monitoring unit to accurately identify real collision characteristics; the shutdown protection module verifies events through the threshold judgment unit, instantly cuts off the power supply through the shutdown trigger unit, and accurately locates the faulty axis through the error output unit, achieving a complete interruption from collision warning to handling. 3. Due to the load-temperature dual-variable compensation technology of the integrated environmental adaptation module, a differential geometric mapping relationship between the thermal deformation field and current temperature drift is constructed through an unsteady heat transfer compensator. The temperature compensation parameters are dynamically adjusted using a load adaptive compensation unit, achieving deep suppression of environmental interference. Combined with the Lyapunov exponential correction mechanism to compress the temperature abrupt change hysteresis window, a deep reinforcement learning algorithm is used to decouple the thermal deformation interference caused by load fluctuations online, synchronously updating the solution logic of the heat conduction equation. This technology overcomes the theoretical limitations of traditional devices under variable temperature load conditions, eliminates the risk of misjudgment caused by current baseline drift, and provides all-weather adaptive protection for wire EDM machines. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a module architecture diagram of the automatic anti-collision device for the lower arm of the slow wire EDM provided in the embodiments of this application; Figure 2 This is a unit architecture diagram of the automatic anti-collision device for the lower arm of the slow wire EDM provided in the embodiments of this application; Figure 3 This is a flowchart of the dynamic adjustment mechanism of the learning rate of the adaptive learning unit provided in the embodiments of this application; Figure 4 This is a flowchart of the reinforcement learning framework of the adaptive learning unit provided in the embodiments of this application; Figure 5 This is a unit architecture diagram of an automatic anti-collision device for the lower arm of a slow wire EDM machine, provided in another embodiment of this application. Figure 6 This is a flowchart of the automatic anti-collision method for the lower arm of a slow wire EDM machine provided in the embodiments of this application.

[0027] Reference numerals: 1. Slow wire EDM system module; 11. Control command unit; 12. Real-time monitoring unit; 13. Adaptive learning unit; 2. Servo drive module; 21. Command parsing unit; 22. Current feedback unit; 23. Noise filtering unit; 3. Linear motor module; 31. Motor drive unit; 32. Position sensing unit; 33. Current generation unit; 4. Stop protection module; 41. Threshold judgment unit; 42. Stop trigger unit; 43. Error output unit; 5. Environmental adaptation module; 51. Temperature sensing unit; 52. Temperature compensation unit; 53. Heat transfer modeling unit; 54. Load adaptive compensation unit. Detailed Implementation

[0028] The following is in conjunction with the appendix Figure 1 -Appendix Figure 6 This application will be described in further detail.

[0029] This application discloses an automatic anti-collision device and method for a slow wire EDM lower arm.

[0030] See attached document Figure 1 An automatic anti-collision device for the lower arm of a slow wire EDM machine includes a slow wire EDM system module 1, a servo drive module 2, a linear motor module 3 containing multiple feed motion axes, and a shutdown protection module 4.

[0031] The slow wire EDM system module 1 is electrically connected to the servo drive module 2. The servo drive module 2 is electrically connected to the linear motor module 3. The linear motor module 3 feeds back the motor current signal to the servo drive module 2. The servo drive module 2 transmits the current waveform curve obtained from the motor current signal to the slow wire EDM system module 1. The slow wire EDM system module 1 is electrically connected to the shutdown protection module 4.

[0032] See attached document Figure 2 The wire EDM system module 1 includes a control command unit 11, a real-time monitoring unit 12, and an adaptive learning unit 13. The servo drive module 2 includes a command parsing unit 21, a current feedback unit 22, and a noise filtering unit 23. The linear motor module 3 includes a motor drive unit 31, a position sensing unit 32, and a current generation unit 33. The shutdown protection module 4 includes a threshold judgment unit 41, a shutdown trigger unit 42, and an error output unit 43.

[0033] The slow wire EDM system module 1 first generates drive commands through the control command unit 11 and sends them to the command parsing unit 21 of the servo drive module 2 for reception and parsing. The command parsing unit 21 of the servo drive module 2 converts the drive commands into motor control signals and sends them to the motor drive unit 31 of the linear motor module 3 to drive the feed motion axis to move. The position sensing unit 32 of the linear motor module 3 detects the actual displacement of the feed motion axis and generates a position feedback signal. The current generation unit 33 of the linear motor module 3 converts the position feedback signal into a motor current signal and feeds it back to the current feedback unit 22 of the servo drive module 2. The current feedback unit 22 of the servo drive module 2 collects the motor current signal in real time to form a current waveform curve. After the noise filtering unit 23 of the servo drive module 2 applies digital filtering technology to eliminate environmental electromagnetic interference, the filtered current waveform curve is transmitted to the real-time monitoring unit 12 of the slow wire EDM system module 1 to monitor the real-time current of each feed motion axis. The real-time monitoring unit 12 of the slow wire EDM system module 1 continuously compares the standard current threshold and standard time threshold for standby shutdown. When the real-time current of any feed axis exceeds the standard current threshold and the duration exceeds the standard time threshold, an alarm is sent to the threshold judgment unit 41 of the shutdown protection module 4. At the same time, the adaptive learning unit 13 of the slow wire EDM system module 1 runs an embedded machine learning algorithm to dynamically adjust the standard current threshold and standard time threshold based on the machine tool's historical current data and real-time machine tool status. The threshold judgment unit 41 of the shutdown protection module 4 confirms the current overrun event, triggers the shutdown trigger unit 42 to immediately cut off the power supply to the linear motor module 3 and lock the corresponding feed axis. The error output unit 43 generates detailed error information containing the feed axis corresponding to the current overrun event and displays it through the human-machine interface.

[0034] When the anti-collision self-learning program is executed, the slow wire EDM system module 1 drives the servo drive module 2 to move each feed motion axis, and reads the peak current and rated current of each feed motion axis during normal axis movement through the current waveform curve fed back by the servo drive module 2, so as to calculate and save the standard current threshold and standard time threshold for each feed motion axis to stop in case of collision.

[0035] The slow wire EDM system module 1 is located in the control core of the machine tool. It is used to issue drive commands and monitor in real time whether the real-time current of each feed axis exceeds the saved standard current threshold and whether its duration exceeds the saved standard time threshold during machine tool use. It is also used to execute the anti-collision self-learning program, which includes an adaptive learning mechanism to dynamically adjust the standard current threshold and standard time threshold according to the machine tool's historical current data and real-time machine tool status.

[0036] In this embodiment, the wire EDM system module 1 includes a control command unit 11, a real-time monitoring unit 12, and an adaptive learning unit 13.

[0037] The control command unit 11, used to generate drive commands, consists of an embedded processor, a servo drive interface card, and a communication module. Its function is to generate precise drive commands to direct the servo drive module 2 to control the movement of each axis of the machine tool, specifically the XYZUV axes in this embodiment, ensuring the safety and efficiency of wire threading, workpiece clamping, and machining processes. When the control command unit 11 receives anti-collision self-learning or axis shifting commands from the wire EDM machine tool system, it parses and generates pulse signals or digital commands through the embedded processor, which are then sent to the servo driver via the servo drive interface card. This drives the linear motor module 3 to execute the corresponding actions, thereby avoiding the risk of collisions caused by human error.

[0038] The real-time monitoring unit 12, used to monitor the real-time current of each feed axis and perform current threshold comparison, consists of a current sensor, an analog-to-digital converter, and a comparison circuit. It monitors the current value of each feed axis in real time and compares it with stored standard current thresholds and duration thresholds to ensure timely intervention in the event of abnormal current caused by a collision. The current sensor continuously collects the linear motor current signal, which is converted into digital data by the analog-to-digital converter. The comparison circuit then compares this data with stored standard current thresholds and standard time thresholds based on peak current and rated current. If the real-time current exceeds the standard time threshold and the duration meets the standard, an error-reporting and shutdown mechanism is immediately triggered.

[0039] The adaptive learning unit 13 is used to run embedded machine learning algorithms to analyze historical current data and real-time status of the machine tool to dynamically adjust and optimize standard current and time thresholds. It consists of a microcontroller, data storage, and embedded machine learning algorithm modules, such as regression analysis models based on historical data.

[0040] Specifically, when the wire EDM system module 1 starts the anti-collision self-learning program, it first sends a drive command to the servo drive module 2, which in turn controls the linear motor module 3 to move each feed axis. During this process, the servo drive module 2 feeds back the current waveform curves of each feed axis to the wire EDM system module 1 in real time. The adaptive learning unit 13 reads and analyzes these current waveform curves through the microcontroller, extracting the peak current and rated current of each feed axis under normal axis movement. At the same time, the microcontroller accesses the historical current data of the machine tool stored in the data memory, including current records from previous machining processes, collision event data, and real-time acquired machine tool status information, including but not limited to the current material type, machining stage, and ambient temperature. The embedded machine learning algorithm module employs a regression analysis model based on historical data to comprehensively analyze the input data. This model fits the trends of historical current data, for example, using the least squares method to establish a linear or nonlinear relationship between current and processing parameters, and combines real-time conditions, such as changes in material hardness or processing stage transitions, to predict reference threshold parameters. The regression analysis model outputs dynamically optimized standard current and time thresholds, ensuring that these thresholds adapt to changes in machine tool operating conditions, such as increasing threshold sensitivity when material hardness increases. After calculation, the microcontroller saves the optimized standard current and time thresholds to the data storage for subsequent monitoring by the shutdown protection module 4. Throughout the process, the adaptive learning mechanism iterates in real time. Each time the anti-collision self-learning program runs, the regression analysis model is retrained based on the newly fed-back current waveform curve and historical dataset, updating the regression analysis model coefficients to compress errors. A reverse verification mechanism is used, for example, substituting the new standard current and time thresholds into historical error event simulations to ensure the accuracy of the standard current and time threshold adjustments. This maintains low false alarm and false alarm rates while improving the robustness and response efficiency of the anti-collision device.

[0041] It should be noted that the regression analysis model is embedded in the microcontroller firmware. When processing data, real-time filtering technology is used to eliminate noise, and the data storage cache mechanism is used to achieve high-speed data read and write. The adaptive adjustment principle relies on the generalization ability of the machine learning model. Through feature engineering, such as extracting current fluctuation statistics and mapping them to the threshold space, the differential correlation between the threshold and the machine tool state is realized.

[0042] Servo drive module 2 is connected to wire EDM system module 1. It is used to receive drive commands from wire EDM system module 1 and drive linear motor module 3 to move. At the same time, it provides feedback on the current waveform curves of the linear motors of each feed motion axis.

[0043] In this embodiment, the servo drive module 2 includes an instruction parsing unit 21, a current feedback unit 22, and a noise filtering unit 23.

[0044] The instruction parsing unit 21, used to convert the drive instructions from the wire EDM system module 1 into motor control signals, consists of a microcontroller, a digital signal processor, and a communication interface. Its function is to convert the drive instructions, typically digital control signals, issued by the wire EDM system module 1 into motor control signals suitable for the linear motor module 3, such as PWM pulse width modulation signals. After receiving the drive instructions, the microcontroller decodes the instructions and maps the signals through the digital signal processor to generate precise motor control signals, ensuring that the drive instructions match the motion parameters of the linear motor, such as speed and acceleration, thereby achieving precise drive of the feed axis.

[0045] The current feedback unit 22, used to acquire the current waveform curve of the linear motor module 3 in real time and transmit it to the wire EDM system module 1, consists of a Hall effect current sensor, an analog-to-digital converter, and a data acquisition card. The Hall effect sensor detects the real-time current of the linear motor and generates an analog voltage signal; the analog-to-digital converter converts the analog signal into a digital signal; the data acquisition card samples and buffers the current waveform to ensure high-precision acquisition and real-time feedback of the current waveform curve, supporting the real-time monitoring function of the wire EDM system module 1.

[0046] The noise filtering unit 23, used to eliminate the influence of environmental electromagnetic interference on the current waveform curve using digital filtering technology, consists of a digital signal processor and an embedded filtering algorithm module, such as an FIR (Finite Impulse Response) filter. After receiving the raw current waveform data from the current feedback unit 22, the digital signal processor runs the embedded filtering algorithm, such as Kalman filtering or moving average filtering, to analyze and remove noise components, such as high-frequency electromagnetic interference, in real time, and outputs the filtered current waveform curve to improve the overall anti-interference capability of the servo drive module 2.

[0047] Linear motor module 3, connected to servo drive module 2, includes linear motors for each feed motion axis, used to realize the movement of each feed motion axis, generate motor current signals and feed them back through servo drive module 2.

[0048] In this embodiment, the linear motor module 3 includes a motor drive unit 31, a position sensing unit 32, and a current generation unit 33.

[0049] The motor drive unit 31, used to drive the feed axes to move according to the motor control signals from the servo drive module 2, consists of a multi-axis control chip such as an FPGA programmable logic device, independent power amplifier modules with IGBT / MOSFET drive circuits configured for each of the XYZUV axes, and inter-axis isolation circuits. It receives multiple motor control signals from the servo drive module 2 and independently drives the linear motor coils of the five feed axes (XYZUV), achieving precise time-sharing / parallel movement of each axis. After parsing the multi-axis control commands, the FPGA multi-axis control chip distributes them to the power amplifiers of each axis through the inter-axis isolation circuits. Each axis power amplifier generates a customized current based on command parameters, such as U-axis acceleration and V-axis velocity, to drive the stator windings of the corresponding linear motor, ensuring the independence and synchronization of electromagnetic thrust in multi-axis coordinated motion.

[0050] The position sensing unit 32 is used to detect the actual displacement of the feed motion axes and generate position feedback signals. It consists of a multi-channel linear encoder system and a dedicated axis signal processor with a digital-to-analog converter and an ARM microprocessor for each axis. In the multi-channel linear encoder system, optical encoders are used for the XYZ axes, and magnetic encoders are used for the UV axes. The actual displacement of the five feed motion axes XYZUV is detected synchronously, generating five independent high-precision position feedback signals to support multi-axis closed-loop control. Each axis encoder acquires displacement data in real time, such as X-axis optical scale pulses and V-axis magnetic grating signals. These data are then denoised, calibrated, and digitized by the dedicated axis processor to output standardized five-axis position signals, ensuring that the wire EDM system module 1 can dynamically calibrate multi-axis trajectory deviations.

[0051] The current generation unit 33 is used to convert the position feedback signals of the XYZUV five axes into motor current signals and feed them back to the servo drive module 2. It consists of a multi-channel Hall current sensor array (one sensor for each feed motion axis), a multi-channel signal conversion module integrating operational amplifiers and digital signal processors, and an axis current fusion processor. Each axis Hall current sensor generates a raw analog current quantity based on the position feedback signal. After conversion, amplification, and filtering, the digital signal processor maps it into a digital current waveform according to the axis number. The axis current fusion processor performs timing alignment and noise suppression on the five signals and outputs an anti-interference multi-axis current dataset, providing a criterion basis for the shutdown protection module 4.

[0052] The shutdown protection module 4 is connected to the slow wire EDM system module 1. It is used to trigger a shutdown and output an error message when the real-time current of any feed motion axis exceeds the standard current threshold and its duration exceeds the standard time threshold.

[0053] In this embodiment, the shutdown protection module 4 includes a threshold judgment unit 41, a shutdown trigger unit 42, and an error output unit 43.

[0054] The threshold judgment unit 41 is used to receive the monitoring results of the slow wire EDM system module 1 and confirm current overrun events. It consists of a high-speed comparator circuit, such as an LM393 chip, a microcontroller, and a multi-axis current data interface supporting XYZUV feed motion axes. It receives real-time monitoring data from the slow wire EDM system module 1, including the current value and duration of each feed motion axis. A preset algorithm is used to confirm whether a current overrun event has occurred, i.e., the real-time current exceeds the standard current threshold and the duration exceeds the standard time threshold. The working principle is based on a threshold comparison mechanism: the microcontroller continuously parses the input data and uses the comparator circuit to compare the current of each axis in real time; when the current overrun conditions of any feed motion axis, such as the X-axis or V-axis, are simultaneously met, a logic judgment signal is triggered, and an event confirmation command is output to the stop trigger unit 42, ensuring the real-time nature of event detection and multi-axis compatibility.

[0055] The shutdown trigger unit 42 is used to immediately cut off the power supply to the linear motor module 3 and lock the feed motion axis corresponding to the current over-limit event upon confirmation of an over-limit event. It consists of a solid-state relay module, an independently configured axis-locking electromagnetic brake for each feed motion axis, and a power cut-off control circuit based on MOSFET switches. After receiving the event command, the relay module disconnects the motor drive power supply through the power cut-off control circuit; simultaneously, the electromagnetic brake applies a physical locking force to the moving parts of the over-limit axis, freezing the axis position and ensuring the safe isolation of the multi-axis system.

[0056] Error output unit 43 is used to generate detailed error information for the feed motion axis corresponding to the current overrun event, including the overrun current value, duration, and feed motion axis number, and displays it through a human-machine interface (HMI). This HMI consists of an embedded processor, an HMI interface module such as an HDMI or TFT-LCD display, and an event log storage module such as an SD card module. The embedded processor obtains event data from threshold judgment unit 41, runs embedded algorithms such as JSON data encapsulation, and generates structured error content. It drives the display screen to output a visual alarm through the interface, and simultaneously stores the event log to support subsequent diagnosis, improving the traceability of multi-axis fault management.

[0057] During the execution of the anti-collision self-learning program, the slow wire EDM system module 1 drives the servo drive module 2 to move each feed motion axis. The peak current and rated current of each feed motion axis during normal axis movement are read through the current waveform curve fed back by the servo drive module 2. The standard current threshold and standard time threshold for each feed motion axis to stop in case of collision are calculated and saved.

[0058] In this embodiment, when executing the anti-collision self-learning program, the wire EDM system module 1 first starts the adaptive learning mechanism and runs a machine learning algorithm, namely, regression analysis or neural network model based on historical current data and real-time machine tool status, to initialize the program and generate drive commands. The control command unit 11 of the wire EDM system module 1 then sends the drive command to the servo drive module 2. The command parsing unit 21 of the servo drive module 2 receives the command and converts it into a precise motor control signal. At the same time, the noise filtering unit 23 of the servo drive module 2 uses digital filtering technology to eliminate the noise influence of environmental electromagnetic interference on the current signal in real time, ensuring the purity of the current waveform curve data. Next, the servo drive module 2 drives the linear motor module 3. The motor drive unit 31 of the linear motor module 3 controls the linear motor movement of each feed motion axis according to the motor control signal, so that the feed motion axis executes a preset movement path, such as uniform speed or acceleration. At the same time, the current generation unit 33 of the linear motor module 3 generates a motor current signal, and the actual displacement of the feed motion axis is detected by the position sensing unit 32 to generate a position feedback signal. This signal is converted into a current waveform curve and is collected in real time by the current feedback unit 22 of the servo drive module 2 and fed back to the slow wire EDM system module 1. After receiving the feedback current waveform curve, the real-time monitoring unit 12 of the slow wire EDM system module 1 analyzes the current waveform curve data and reads the peak current (i.e., the maximum instantaneous current value during the movement) and the rated current (i.e., the stable current value) of each feed motion axis during the normal axis shifting process. The average current value under working conditions, combined with the machine tool's historical current database storing past operating data and real-time status parameters such as temperature and load, is dynamically analyzed by the adaptive learning unit 13. The machine learning algorithm determines the standard current threshold for each feed axis, i.e., the upper limit of the current that triggers shutdown, by iteratively calculating, for example, by multiplying the peak current by a safety factor and considering the fluctuation range of the rated current. At the same time, based on the duration statistics of current overrun events, such as the average response time of historical collision events, the standard time threshold is calculated, i.e., the upper limit of the allowable duration after current overrun. Finally, the slow wire EDM system module 1 saves the calculated standard current threshold and standard time threshold to the internal memory for subsequent real-time monitoring and for the threshold judgment unit 41 of the shutdown protection module 4 to call. The whole process achieves self-optimization under closed-loop control.

[0059] The following is a description of another embodiment of the method and system provided in this application.

[0060] Based on the automatic anti-collision device for the lower arm of a slow wire EDM machine in Embodiment 1, this Embodiment 2 adds some specific implementation methods.

[0061] See attached document Figure 3In this embodiment, the adaptive learning unit 13 also integrates a dynamic learning rate adjustment mechanism. This mechanism adjusts the learning rate of the embedded machine learning algorithm in real time based on the differences in sensitivity to current fluctuations between the type of processed material and the processing stage. Specifically, when the hardness level of the processed material increases or the processing stage changes, the learning rate automatically increases to accelerate the updating of the standard current threshold and the standard time threshold; when the hardness level of the processed material decreases or the processing stage is in a stable cutting state, the learning rate automatically decreases to suppress noise interference.

[0062] The dynamic learning rate adjustment mechanism refers to the software control logic embedded in the adaptive learning unit 13, which dynamically modifies the learning rate parameter of the machine learning model based on external input parameters to optimize the threshold update efficiency. The processing material type refers to the type of workpiece material currently being processed by the wire EDM machine, including metals, alloys, or composite materials, whose properties are preset by the machine tool control core or obtained through real-time sensors. The processing stage refers to the temporal state during processing, including initial entry, stable cutting, finishing, and final stages, which are divided by the real-time monitoring unit 12 of the wire EDM system module 1 based on the current waveform curve and position feedback signal. The hardness grade refers to the mechanical hardness quantification index of the processed material type, based on international standards such as Rockwell hardness HRC or Brinell hardness HB grading; a higher value indicates a harder material. Stable cutting refers to the state where the current fluctuation amplitude is below the preset variance threshold during the processing stage, at which point the cutting process is smooth and noise interference is minimal. The wire EDM system module 1 drives this process when executing the adaptive learning program. The adaptive learning unit 13 implements the dynamic learning rate adjustment mechanism through built-in embedded machine learning algorithms, such as stochastic gradient descent or a variant of the Adam optimizer.

[0063] Specifically, the system first obtains hardness grade data of the processed material type from the real-time monitoring unit 12, for example, through matching with a preset material library or input from an online hardness sensor and processing stage information, such as by analyzing the frequency characteristics and variance of the current waveform curve fed back by the servo drive module 2. When the variance is below 0.5A² for five consecutive sampling periods, it is determined to be in a stable cutting stage. Then, the learning rate dynamic adjustment mechanism calculates the learning rate adjustment coefficient based on a sensitivity difference model. This sensitivity difference model is a pre-trained multilayer perceptron neural network, with the input being the hardness grade and processing stage encoding, such as a one-hot vector, and the output being the learning rate scaling factor. When the hardness grade increases, such as when the HRC (Rockwell C scale) value increases by 10% or the processing stage changes, such as switching from stable cutting to finishing, the scaling factor automatically increases to 1.5-2.0 times to accelerate threshold updates; when the hardness grade decreases, such as when the HRC value decreases by 15%, or when the processing stage is in stable cutting, the scaling factor automatically decreases to 0.5-0.8 times to suppress noise caused by environmental electromagnetic interference. During the adjustment process, the adaptive learning unit 13 applies a scaling factor to the learning rate parameter of the embedded machine learning algorithm, such as an initial learning rate of 0.001, and iteratively updates the standard current threshold and standard time threshold through gradient descent.

[0064] It should be noted that this dynamic learning rate adjustment mechanism achieves real-time response at the hardware level through an FPGA or microcontroller, while at the software level, a control loop written in C++ samples data every 100 milliseconds to ensure a lag time of less than 10ms. Furthermore, this dynamic learning rate adjustment mechanism integrates an anomaly handling submodule. When a sudden change in the machining stage is detected, such as a sudden change in hardness grade exceeding 20%, the learning rate upper limit protection is automatically triggered, for example, not exceeding three times the initial value, to prevent overfitting. Ultimately, this design effectively reduces the false alarm rate by dynamically compressing the noise interference window (during the stable cutting stage) and accelerating learning (during the material hardness change stage), and ensures that threshold updates are synchronized with the real-time machine tool status, avoiding false triggering or missed alarms of the anti-collision device.

[0065] See attached document Figure 4 In this embodiment, the adaptive learning unit 13 adopts a reinforcement learning framework based on meta-learning, which includes a dynamic feature extraction layer and a policy transfer layer.

[0066] The dynamic feature extraction layer uses a multi-scale convolutional neural network to simultaneously process the trend features of historical current data of the machine tool and the transient fluctuation features of the machine tool's real-time status, generating a current pattern vector that integrates spatiotemporal characteristics. The policy transfer layer constructs a digital twin of the machine tool's operation based on the current pattern vector, and pre-simulates the collision risk evolution path under different threshold combinations in the digital twin, selecting the reference threshold combination that minimizes the sum of the false alarm rate and the false negative rate.

[0067] The adaptive learning unit 13 initiates a reverse verification mechanism after each update of the reference threshold combination. It substitutes the current reference threshold combination into the historical error event data for virtual replay. Only when the matching degree between the virtual replay result and the actual error event exceeds the preset verification standard will the reference threshold combination be applied to dynamic adjustment and optimization.

[0068] The reinforcement learning framework based on meta-learning refers to an algorithmic architecture that combines meta-learning and reinforcement learning for efficiently transferring and optimizing collision avoidance strategies. The dynamic feature extraction layer is the neural network processing module within this framework, responsible for extracting multi-scale features from raw current data. The policy transfer layer is the decision optimization module within the framework, used to transfer the learned policy to a digital twin for simulation. The digital twin is a virtual simulation model of the machine tool's operation, mirroring the behavior of the physical system through real-time data. The reverse verification mechanism is a threshold verification procedure used to ensure the reliability of the updated threshold combination in practical applications. The false alarm rate is the proportion of collision avoidance devices that erroneously trigger shutdown, and the false negative rate is the proportion of real collision events that are not detected in time; the sum of these two is used to quantify collision avoidance performance. The wire EDM system module 1 drives this process when executing the adaptive learning program, and the adaptive learning unit 13 implements this reinforcement learning framework through embedded hardware, such as an FPGA or an industrial-grade GPU.

[0069] Specifically, the dynamic feature extraction layer employs a multi-scale convolutional neural network, such as a custom architecture based on ResNet-18. Its input includes historical machine tool current data, such as the current waveform curve over the past 72 hours, with a sampling frequency of 1kHz, capturing trend characteristics. Long-term trends are extracted through time series analysis, such as ARIMA models or Fourier transforms, calculating moving averages and slopes. The input also includes real-time machine tool status, such as the current value of the current feed axis, temperature sensor data, and transient fluctuations in load feedback. High-frequency noise and sudden peaks are extracted through short-time Fourier transforms or wavelet transforms. The network structure is designed with parallel convolutional paths, with kernel sizes of 3x3, 5x5, and 7x7 to simultaneously capture features at different time scales. The output layer uses a fully connected network to fuse spatiotemporal characteristics to generate a current pattern vector, such as a 256-dimensional feature vector containing mean, variance, and spectrum. Energy, etc.; the policy transfer layer, based on this current pattern vector, uses meta-learning algorithms, such as Model-Agnostic Meta-Learning, to construct a digital twin. The construction process can be implemented in Unity3D or MATLAB Simulink. This digital twin integrates a machine tool dynamics model, including the inertia and friction parameters of linear motor module 3. In a virtual environment, it pre-simulates the collision risk evolution path under different standard current thresholds, ranging from 0.5A to 3.0A with a step size of 0.05A, and standard time thresholds, ranging from 5ms to 200ms with a step size of 5ms. It runs 10,000 iterations through Monte Carlo simulation, calculates the false alarm rate and false negative rate for each combination, and applies optimization algorithms, such as genetic algorithms or Bayesian optimization, to select the reference threshold combination that minimizes the sum of the two, for example, selecting the combination that makes the sum less than 5%.

[0070] It should be noted that the reverse verification mechanism is automatically triggered after each update of the reference threshold combination. It calls the historical error event database via a data interface, storing all error records from the past 90 days, including timestamps, current values, and event types. This is then used in a digital twin to virtually replay the event, simulating the application of the new threshold combination to historical data and generating a virtual error sequence. The matching degree is calculated based on confusion matrix metrics such as precision, recall, and F1 score. The preset verification standard is an F1 score exceeding 0.95. Only when this is met will the adaptive learning unit 13 dynamically update the reference threshold combination to the real-time monitoring unit 12. Furthermore, the mechanism incorporates fault-tolerant processing; for example, when the matching degree falls below 0.9, it automatically rolls back to the previous valid threshold. A real-time operating system, such as VxWorks, ensures a processing latency of less than 20ms, preventing overfitting and system oscillations. Ultimately, this reduces the total number of false alarms and false negatives of the collision avoidance device by more than 18% during testing, providing sufficient technical support.

[0071] See attached document Figure 5In this embodiment, the automatic anti-collision device for the lower arm of the wire EDM machine also includes an environmental adaptation module 5, which is connected to the linear motor module 3 and the servo drive module 2. It is used to collect ambient temperature data in real time and correct the current waveform curve through a temperature compensation algorithm. At the same time, it embeds an unsteady heat transfer compensator to construct a differential geometric mapping relationship between the machine tool thermal deformation field and the current temperature drift. By solving the divergence term in the heat conduction equation in real time, it separates the current baseline drift caused by temperature. The temperature compensation algorithm adopts the Lyapunov exponential correction mechanism to dynamically compress the current monitoring lag window caused by sudden changes in ambient temperature.

[0072] The environmental adaptation module 5 includes a temperature sensing unit 51, a temperature compensation unit 52, and a heat transfer modeling unit 53.

[0073] The temperature sensing unit 51 is used to collect ambient temperature data in real time for each feed axis area of ​​the machine tool, and convert the analog temperature signal into a digital signal for transmission to the temperature compensation unit 52. It consists of a multi-channel platinum resistance temperature sensor array covering the XYZUV feed axis areas, such as a PT100; a signal conditioning circuit with an instrumentation amplifier; and an analog-to-digital converter, such as an ADS1248. The multi-channel platinum resistance sensor array is deployed at key points in the linear motor module 3, such as the motor coil and guide rail, to detect changes in ambient temperature. The signal conditioning circuit amplifies and linearizes the weak resistance signal. The analog-to-digital converter digitizes the temperature data at a 1kHz sampling rate, ensuring an accuracy of ±0.1℃ and providing dynamic input for the temperature compensation algorithm.

[0074] The temperature compensation unit 52 is used to dynamically adjust the temperature compensation algorithm through the Lyapunov exponent correction mechanism, correct the current waveform curve fed back by the servo drive module 2, and compress the monitoring lag window caused by temperature abrupt changes. It consists of an embedded processor, firmware integrating the temperature compensation algorithm with the Lyapunov exponent correction mechanism, and non-volatile memory storing compensation parameters. After receiving data from the temperature sensing unit 51, the embedded processor runs the embedded algorithm to calculate the current temperature drift offset based on the real-time temperature, separating the baseline drift caused by temperature from the original current waveform curve. It uses the Lyapunov exponent to evaluate the temperature abrupt change rate and adaptively shortens the current monitoring response time, for example, from 100ms to 20ms. The purified current waveform curve is fed back to the slow wire EDM system module 1 to ensure that real-time monitoring is not affected by temperature drift.

[0075] The heat transfer modeling unit 53 is used to construct the differential geometric mapping relationship between the machine tool's thermal deformation field and the current temperature drift. It predicts the temperature drift trend by solving the divergence term of the heat conduction equation. It consists of a multi-core digital signal processor, a thermodynamic computation accelerator, and a temperature-current mapping database. The thermodynamic computation accelerator is used to implement finite element analysis. The multi-core digital signal processor trains a nonlinear mapping model based on historical temperature-current data, such as a manifold learning algorithm, to establish the mathematical relationship between the thermal deformation field of the machine tool structure and the current temperature drift. The thermodynamic computation accelerator discretizes the three-dimensional heat conduction equation, separating the divergence component caused by the temperature gradient; then, it outputs the thermal deformation field mapping parameters to the temperature compensation unit 52 to predict the temperature drift direction within the next 5 seconds, achieving advance compensation.

[0076] Specifically, the unsteady-state heat transfer compensator refers to an embedded software module developed based on the first law of thermodynamics, which discretizes the machine tool structure using the finite element method. The differential geometric mapping relationship refers to a mathematical model describing the nonlinear relationship between the thermal deformation field (geometric deformation of the machine tool due to temperature gradients) and current temperature drift (current deviation due to changes in motor coil resistance with temperature). The Lyapunov exponential correction mechanism refers to an adaptive control algorithm designed using Lyapunov stability theory to optimize temperature compensation parameters in real time; the current monitoring hysteresis window refers to the time delay between a sudden change in ambient temperature and the effective application of current correction.

[0077] In practical implementation, ambient temperature data is first collected in real time using PT100 platinum resistance temperature sensors distributed on the stator side of the linear motor at a sampling frequency of 1kHz. The raw temperature data is then filtered by a Kalman filter to eliminate measurement noise before being input into the temperature compensation algorithm. This temperature compensation algorithm employs a dual closed-loop control structure. The inner loop constructs the machine tool thermal deformation field through an unsteady heat transfer compensator—discretizing the machine tool structure into 10,000 finite element nodes, solving the unsteady heat conduction equation at each node, and specifically using the explicit Euler method to iteratively calculate the temperature gradient distribution for the divergence term. The outer loop, based on the calculated temperature gradient field, maps the thermal deformation field to the current temperature drift space through a pre-trained differential geometric mapping model implemented by a tensor field network, and outputs the current baseline drift in real time. The Lyapunov exponential correction mechanism dynamically adjusts the compensation parameters—defining the state error e and constructing the Lyapunov function. By ensuring V<0, the compensation gain update law is derived, thereby effectively compressing the lag window.

[0078] In this embodiment, the environmental adaptation module 5 further includes a load adaptive compensation unit 54, which is used to monitor the load change of the linear motor module 3 in real time and dynamically adjust the parameters of the temperature compensation algorithm through a deep deterministic strategy gradient reinforcement learning algorithm to compensate for the additional thermal deformation caused by the load change; the output of the load adaptive compensation unit 54 is used to update the differential geometric mapping relationship.

[0079] Among them, the deep deterministic policy gradient reinforcement learning algorithm refers to a machine learning method that combines deep neural networks and policy optimization, including an Actor network to generate action policies and a Critic network to evaluate state values. Additional thermal deformation refers to the incremental thermal expansion of the machine tool structure caused by load changes, exceeding the basic temperature compensation range. Differential geometric mapping relationship refers to a mathematical model describing the nonlinear correlation between the thermal deformation field and current temperature drift, parameterized through a tensor field.

[0080] The load adaptive compensation unit 54 operates continuously during machine tool processing. Specifically, it first collects load change data in real time, such as axial pressure values ​​(range 0-500N), via a piezoelectric force sensor installed in the linear motor module 3 at a sampling frequency of 500Hz. The raw data, after preprocessing including normalization and Kalman filtering, is input into the state space of the deep deterministic policy gradient reinforcement learning algorithm. The state vector includes the load value, ambient temperature, current waveform curve, and the real-time position of the machine tool. The deep deterministic policy gradient reinforcement learning algorithm framework is designed as a dual-network structure. The Actor network (a three-layer fully connected neural network with 128 hidden nodes) outputs the action policy, i.e., the adjustment amount of the temperature compensation algorithm parameters, such as the gain coefficient and integral time constant. The Critic network (a convolutional neural network) processes the time series, evaluates the state value, and calculates the Q-value reward. The reward function is defined as the negative square of the current monitoring error. An experience replay pool with a capacity of 10,000 samples and Ornstein-Uhlenbeck noise are used to explore and optimize the strategy. The training cycle is one iteration per second, and the learning rate is set to 0.001. The output of the load adaptive compensation unit 54, i.e. the optimized parameters, is used to update the differential geometric mapping relationship in real time—adjusting the thermal strain-current transformation matrix in the mapping model through the online gradient descent method to ensure that the mapping relationship dynamically adapts to the thermal deformation offset caused by the load.

[0081] This application also provides an automatic anti-collision method for the lower arm of a slow wire EDM machine, which can be applied to the device in any of the above embodiments, with reference to... Figure 6 It illustrates a flowchart of a method provided in an embodiment of this application, the method comprising steps S1-S4: S1. The anti-collision self-learning program is executed through the slow wire EDM system module 1 to drive the servo drive module 2 to work and move the linear motor module 3 of each feed motion axis. The peak current and rated current of each feed motion axis during normal axis movement are read through the current data curve fed back by the servo drive module 2.

[0082] S2. Based on the peak current and rated current, calculate and save the standard current threshold and standard time threshold for each feed motion axis.

[0083] S3. During the use of the machine tool, the real-time current of each feed axis is monitored in real time through the slow wire EDM system module 1.

[0084] S4. When the real-time current of any feed axis exceeds the standard current threshold and its duration exceeds the standard time threshold, the shutdown protection module 4 will trigger a shutdown and output an error message.

[0085] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," "third," and similar terms used in this application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. The terms "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "comprising" or "including" and similar terms mean that the elements or objects preceding "comprising" or "including" encompass the elements or objects listed following "comprising" or "including" and their equivalents, and do not exclude other elements or objects. "Above," "below," "left," "right," etc., are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0086] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. An automatic anti-collision device for the lower arm of a slow wire EDM machine tool, characterized in that: It includes a slow wire EDM system module (1), a servo drive module (2), a linear motor module containing multiple feed axes (3), and a shutdown protection module (4); The slow wire EDM system module (1) is located in the control core of the machine tool and is used to issue drive commands and monitor in real time whether the real-time current of each feed motion axis exceeds the saved standard current threshold and whether its duration exceeds the saved standard time threshold during the use of the machine tool. It is also used to execute a collision avoidance self-learning program, which includes an adaptive learning mechanism to dynamically adjust the standard current threshold and the standard time threshold based on the machine tool's historical current data and real-time machine tool status; The servo drive module (2) is connected to the slow wire EDM system module (1) and is used to receive the drive command of the slow wire EDM system module (1) and drive the linear motor module (3) to move, while feeding back the current waveform curve of each feed motion axis linear motor. The linear motor module (3) is connected to the servo drive module (2) and includes linear motors for each feed motion axis. It is used to realize the movement of each feed motion axis, generate motor current signals and feed them back through the servo drive module (2). The shutdown protection module (4) is connected to the slow wire EDM system module (1) and is used to trigger a shutdown and output an error message when the real-time current of any feed motion axis exceeds the standard current threshold and its duration exceeds the standard time threshold. During the execution of the anti-collision self-learning program, the slow wire EDM system module (1) drives the servo drive module (2) to move each feed motion axis. The peak current and rated current of each feed motion axis during normal axis movement are read through the current waveform curve fed back by the servo drive module (2). The standard current threshold and the standard time threshold for each feed motion axis to stop due to collision are calculated and saved.

2. The automatic anti-collision device for the lower arm of a slow wire EDM machine according to claim 1, characterized in that: The slow wire EDM system module (1) includes a control command unit (11), a real-time monitoring unit (12), and an adaptive learning unit (13); The control instruction unit (11) is used to generate drive instructions; The real-time monitoring unit (12) is used to monitor the real-time current of each feed motion axis and perform current threshold comparison. The adaptive learning unit (13) is used to run an embedded machine learning algorithm to analyze the historical current data and real-time status of the machine tool to dynamically adjust and optimize the standard current threshold and the standard time threshold.

3. The automatic anti-collision device for the lower arm of a slow wire EDM machine according to claim 1, characterized in that: The servo drive module (2) includes an instruction parsing unit (21), a current feedback unit (22), and a noise filtering unit (23); The instruction parsing unit (21) is used to convert the drive instructions of the slow wire EDM system module (1) into motor control signals; The current feedback unit (22) is used to collect the current waveform curve of the linear motor module (3) in real time and transmit it to the slow wire EDM system module (1); The noise filtering unit (23) is used to apply digital filtering technology to eliminate the influence of environmental electromagnetic interference on the current waveform curve.

4. The automatic anti-collision device for the lower arm of a slow wire EDM machine according to claim 3, characterized in that: The linear motor module (3) includes a motor drive unit (31), a position sensing unit (32), and a current generation unit (33); The motor drive unit (31) is used to drive the feed motion axis to move according to the motor control signal of the servo drive module (2); The position sensing unit (32) is used to detect the actual displacement of the feed motion axis and generate a position feedback signal; The current generation unit (33) is used to convert the position feedback signal into a motor current signal and feed it back to the servo drive module (2).

5. The automatic anti-collision device for the lower arm of a slow wire EDM machine according to claim 1, characterized in that: The shutdown protection module (4) includes a threshold judgment unit (41), a shutdown trigger unit (42), and an error output unit (43); The threshold judgment unit (41) is used to receive the monitoring results of the slow wire EDM system module (1) and confirm the current exceeding the standard event; The shutdown trigger unit (42) is used to immediately cut off the power supply of the linear motor module (3) and lock the feed motion axis corresponding to the current over-limit event after confirming the current over-limit event; The error output unit (43) is used to generate detailed error information of the feed motion axis corresponding to the current over-limit event and display it through the human-machine interface.

6. The automatic anti-collision device for the lower arm of a slow wire EDM machine according to claim 2, characterized in that: The adaptive learning unit (13) also integrates a dynamic learning rate adjustment mechanism; the dynamic learning rate adjustment mechanism adjusts the learning rate of the embedded machine learning algorithm in real time based on the difference in sensitivity of the processing material type and processing stage to current fluctuations. Specifically, when the hardness grade of the processed material increases or the processing stage changes, the learning rate automatically increases to accelerate the updating of the standard current threshold and the standard time threshold; when the hardness grade of the processed material decreases or the processing stage is in a stable cutting state, the learning rate automatically decreases to suppress noise interference.

7. The automatic anti-collision device for the lower arm of a slow wire EDM machine according to claim 2, characterized in that: The adaptive learning unit (13) adopts a reinforcement learning framework based on meta-learning, which includes a dynamic feature extraction layer and a policy transfer layer. The dynamic feature extraction layer uses a multi-scale convolutional neural network to simultaneously process the trend features of the machine tool's historical current data and the transient fluctuation features of the machine tool's real-time state, generating a current pattern vector that integrates spatiotemporal characteristics. The strategy migration layer constructs a digital twin of the machine tool operation based on the current mode vector, and pre-simulates the collision risk evolution path under different threshold combinations in the digital twin to select the reference threshold combination that minimizes the sum of false alarm rate and false negative rate. The adaptive learning unit (13) initiates a reverse verification mechanism after each update of the reference threshold combination, substitutes the current reference threshold combination into the historical error event data for virtual replay, and applies the reference threshold combination to dynamic adjustment and optimization only when the matching degree between the virtual replay result and the actual error event exceeds the preset verification standard.

8. The automatic anti-collision device for the lower arm of a slow wire EDM machine according to claim 1, characterized in that: It also includes an environmental adaptation module (5), which is connected to the linear motor module (3) and the servo drive module (2) for real-time acquisition of ambient temperature data and correction of the current waveform curve through a temperature compensation algorithm. At the same time, it embeds an unsteady heat transfer compensator to construct the differential geometric mapping relationship between the machine tool thermal deformation field and the current temperature drift. By solving the divergence term in the heat conduction equation in real time, it separates the current baseline drift caused by temperature. The temperature compensation algorithm adopts the Lyapunov exponential correction mechanism to dynamically compress the current monitoring lag window caused by sudden changes in ambient temperature.

9. The automatic anti-collision device for the lower arm of a slow wire EDM machine according to claim 8, characterized in that: The environmental adaptation module (5) further includes a load adaptive compensation unit (54), which is used to monitor the load change of the linear motor module (3) in real time and dynamically adjust the parameters of the temperature compensation algorithm through a deep deterministic strategy gradient reinforcement learning algorithm to compensate for the additional thermal deformation caused by the load change; the output of the load adaptive compensation unit (54) is used to update the differential geometric mapping relationship.

10. An automatic anti-collision method for the lower arm of a slow wire EDM machine, as described in any one of claims 1 to 9, characterized in that: The method includes S1, executing an anti-collision self-learning program through the slow wire EDM system module (1), driving the servo drive module (2) to work to move the linear motor module (3) of each feed motion axis, and reading the peak current and rated current of each feed motion axis during normal axis shifting through the current data curve fed back by the servo drive module (2). S2. Based on the peak current and the rated current, calculate and save the standard current threshold and standard time threshold for each feed motion axis. S3. During the use of the machine tool, the real-time current of each feed axis is monitored in real time through the slow wire EDM system module (1); S4. When the real-time current of any feed axis exceeds the standard current threshold and its duration exceeds the standard time threshold, the shutdown protection module (4) triggers a shutdown and outputs an error message.

Images