A wire cutting machine feed speed control system and method

Abstract

The present invention relates to the field of cutting machine control technology, specifically a wire cutting machine feed speed control system and method, including: a multimodal signal identification material module uses adaptive feature selection technology to extract current and vibration signal features, based on an improved random forest model, combined with dynamic weight voting and weighted fusion method to identify processing materials, and supports online learning and model updates to meet the processing requirements of different materials. The dual-loop adaptive control module includes outer loop initial compensation and inner loop fuzzy PID adjustment. According to historical processing records and dual-target error values, it optimizes real-time fine-tuning parameters to ensure processing accuracy and stability, and improves processing quality. The energy consumption management optimization module intelligently regulates system energy consumption by identifying non-processing and processing time periods, improves overall energy efficiency, and achieves energy-saving production. The present invention improves the efficiency and accuracy of wire cutting machines in multi-variety small-batch mold processing.

Description

Technical Field

[0001] The invention relates to the technical field of cutting machine control, and in particular to a feed speed control system and method for a wire cutting machine. Background Art

[0002] As the core technology of precision mold manufacturing and complex parts processing, wire cutting can efficiently process high-hardness conductive materials and multi-variety small-batch production tasks. However, the current mainstream wire cutting machine control system has the following problems:

[0003] First, existing technologies mostly use open-loop or semi-closed-loop control architectures and rely on preset process parameter libraries to operate. When faced with scenarios where material types and workpiece thicknesses change frequently, their ability to quickly change models is insufficient and their material recognition accuracy is limited. Secondly, due to the lag in the adjustment of the feed speed, when switching between multiple materials, it is very easy to cause unstable processing accuracy, especially for workpieces with complex contours. During high-speed cutting, errors continue to accumulate, which may affect the final processing quality. Finally, these technologies still maintain a high level of energy consumption when they are idle and in standby mode. They fail to effectively optimize the energy utilization rate during the processing process, and lack energy efficiency optimization strategies for different materials and different process parameters, which increases the operating cost of the processing process.

[0004] In order to improve the efficiency and accuracy of wire cutting machine in processing multi-variety and small batch molds, a wire cutting machine feed speed control system and method are proposed. Summary of the invention

[0005] The purpose of the present invention is to provide a feed speed control system and method for a wire cutting machine, which uses adaptive feature selection technology to extract current and vibration signal features through a multimodal signal recognition material module, identifies processing materials based on an improved random forest model, combined with dynamic weight voting and weighted fusion methods, and supports online learning and model updates to meet the processing requirements of different materials. The dual-loop adaptive control module includes outer loop initial compensation and inner loop fuzzy PID adjustment. According to historical processing records and dual-target error values, it optimizes fine-tuning parameters in real time to ensure processing accuracy and stability, thereby improving processing quality. The energy consumption management optimization module intelligently regulates system energy consumption by identifying non-processing and processing periods, improves overall energy efficiency, and achieves energy-saving production. The present invention improves the efficiency and accuracy of wire cutting machines in processing multi-variety small-batch molds.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A wire cutting machine feed speed control system, comprising:

[0008] The multimodal signal material identification module includes: a signal feature extraction unit, which is used to extract material current signal features and material vibration signal features according to an adaptive feature selection mechanism; a material classification unit, which is used to establish an improved random forest model according to the signal feature extraction unit, wherein the random forest model sets a stage-specific subtree set for different processing stages, and uses a dynamic weight voting mechanism and a weighted fusion method to identify processing materials; a model optimization unit, which is used to perform online learning and model update on the random forest model;

[0009] The dual-loop adaptive control module includes: an outer-loop control unit for acquiring basic process parameters for the processing material and performing initial compensation according to historical processing records; an inner-loop control unit for adjusting real-time fine-tuning parameters using a fuzzy PID model according to dual-target error values;

[0010] The energy consumption management optimization module is used to identify non-processing periods and processing periods according to trigger conditions and optimize energy consumption using dynamic control strategies.

[0011] Furthermore, the adaptive feature selection mechanism includes:

[0012] Setting a preset rule for determining the processing stage of the wire cutting machine;

[0013] Establishing a feature-stage correlation matrix for setting feature weights and feature priorities for the material current signal features and the material vibration signal features according to the processing stage using a Pearson correlation coefficient method;

[0014] For the feature-stage correlation matrix, a feature selection model is used to screen material features and update the feature weights and the feature priorities.

[0015] Furthermore, the process of updating the feature weights includes:

[0016] ;

[0017] in, is the feature weight of feature i at time t+1, is the feature weight of the feature i at time t, is the change in classification accuracy of the feature i, is the change in the eigenvalue of the feature i, For the forgetting factor.

[0018] Furthermore, the material current signal characteristics and the material vibration signal characteristics include:

[0019] The material current signal characteristics include: time domain characteristics, including waveform distortion rate, pulse density and peak-to-peak ratio; frequency domain characteristics, including fundamental wave energy ratio, third harmonic amplitude and band energy entropy; time-frequency domain characteristics, including wavelet energy moment and time-frequency ridge slope; statistical characteristics, including kurtosis coefficient and waveform factor;

[0020] The material vibration signal characteristics include: frequency domain analysis characteristics, including main frequency band energy ratio and harmonic wave attenuation rate; time-frequency analysis characteristics, including wavelet packet node energy and instantaneous frequency variance; nonlinear characteristics, including Lyapunov exponent and fractal dimension; statistical characteristics, including zero crossing rate and kurtosis skewness joint factor.

[0021] Furthermore, the model optimization unit includes:

[0022] If there is a new sample, check whether the new sample is misclassified; if misclassified, put the new sample into a buffer pool for periodically updating the stage-specific subtree set;

[0023] If the error rate of the stage-specific subtree set is greater than a preset threshold and the classification accuracy of the stage-specific subtree set decreases by more than M% for N consecutive times, the subtree is replaced.

[0024] Furthermore, the adjustment process of the real-time fine-tuning parameters includes:

[0025] Calculate surface roughness deviation and material removal rate;

[0026] For each of the processing stages, the processing weights of the surface roughness deviation and the material removal rate are allocated, and the dual-objective error value is obtained using a weighted summation method;

[0027] If the dual-objective error value is less than the optimization target value, the real-time fine-tuning parameter is adjusted using the fuzzy PID model; otherwise, the real-time fine-tuning parameter is maintained.

[0028] Furthermore, the fuzzy PID model includes:

[0029] Calculate the change rate of the dual target error values ​​to obtain the material error change rate;

[0030] The material error change rate and the dual-target error value are set as material input variables, and the material proportional coefficient adjustment amount, the material integral coefficient adjustment amount and the material differential coefficient adjustment amount are set as parameter output variables;

[0031] Fuzzifying the material input variable and the parameter output variable to obtain a material fuzzy input variable and a material fuzzy output variable;

[0032] Setting a material fuzzy rule table, for obtaining the material fuzzy output variable according to the material fuzzy input variable;

[0033] The fuzzy output variable is defuzzified using the center of gravity method, and the PID parameters are adjusted to obtain the real-time fine-tuning parameters.

[0034] A method for controlling the feeding speed of a wire cutting machine, comprising:

[0035] According to the adaptive feature selection mechanism, the material current signal features and material vibration signal features are extracted;

[0036] According to the material current signal characteristics and the material vibration signal characteristics, an improved random forest model is established, wherein the random forest model sets stage-specific subtree sets for different processing stages, and uses a dynamic weight voting mechanism and a weighted fusion method to identify processing materials;

[0037] Performing online learning and model updating on the random forest model;

[0038] Retrieving basic process parameters for the processing material and performing initial compensation according to historical processing records;

[0039] According to the dual-objective error value, the real-time fine-tuning parameters are adjusted using the fuzzy PID model;

[0040] Identify non-processing and processing periods based on trigger conditions and use dynamic control strategies to optimize energy consumption.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] 1. The present invention utilizes a multimodal signal material identification module to extract the material current signal and vibration signal features in wire cutting through an adaptive feature selection mechanism. Based on the improved random forest model, a stage-specific subtree set is constructed, and combined with a dynamic weight voting mechanism and a weighted fusion method, the processing material can be accurately identified. This module effectively improves the recognition accuracy of the wire cutting machine under different materials and process stages, enhances the system's adaptability to complex workpiece materials, and improves the stability and consistency of processing.

[0043] 2. The present invention adopts a dual-loop adaptive control module, which retrieves basic process parameters in the outer loop control and performs initial compensation in combination with historical records. In the inner loop control, based on the dual target error value, the fuzzy PID control algorithm is used to dynamically adjust the real-time fine-tuning parameters to ensure that key process parameters such as feed speed and pulse parameters maintain the optimal processing state when switching between multiple materials. This module improves the stability and dimensional accuracy of the processing quality, thereby improving the processing efficiency of the wire cutting machine for multiple varieties and small batch molds.

[0044] 3. The present invention uses an energy consumption management optimization module to intelligently identify non-processing periods and processing periods based on trigger conditions such as discharge signal status, servo motor motion status, process task queue and environmental sensor signals, and adopts dynamic energy consumption optimization strategies, including energy-saving measures such as adaptive adjustment of working fluid flow and pulse power supply gap voltage regulation, thereby reducing no-load energy consumption and invalid power consumption, so that the wire cutting machine can still maintain energy-saving and stable processing performance under long-term operation and complex working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The present invention provides a structural schematic diagram of a feed speed control system for a wire cutting machine;

[0046] Figure 2 A schematic diagram of a process for identifying material categories is provided for the present invention;

[0047] Figure 3 A schematic flow chart of a dual-loop adaptive control module is provided for the present invention;

[0048] Figure 4 The present invention provides a flow chart of a method for controlling the feed speed of a wire cutting machine. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] See also Figures 1 to 4 The present invention provides a wire cutting machine feed speed control system and method, and the technical solution is as follows:

[0051] Embodiment 1:

[0052] With the continuous development of the manufacturing industry, molds play a vital role in product production, especially the growing demand for multi-variety and small-batch mold processing, which puts higher requirements on the efficiency and accuracy of processing equipment. As one of the key equipment for mold processing, the performance of wire cutting machine directly affects the processing quality and production efficiency of the mold. The traditional wire cutting machine feed speed control system has problems such as inaccurate material identification, limited control accuracy and high energy consumption. In order to solve these problems, such as Figure 1 As shown, a wire cutting machine feed speed control system includes:

[0053] The multimodal signal recognition material module includes: a signal feature extraction unit, a material classification unit and a model optimization unit.

[0054] Among them, the signal feature extraction unit is used to extract the material current signal features and the material vibration signal features according to the adaptive feature selection mechanism; the material classification unit is used to establish an improved random forest model according to the signal feature extraction unit, and the random forest model sets a stage-specific subtree set for different processing stages, and uses a dynamic weight voting mechanism and a weighted fusion method to identify processing materials; the model optimization unit is used to perform online learning and model update on the random forest model.

[0055] Furthermore, the adaptive feature selection mechanism includes:

[0056] Setting a preset rule for determining the processing stage of the wire cutting machine;

[0057] Establishing a feature-stage correlation matrix for setting feature weights and feature priorities for the material current signal features and the material vibration signal features according to the processing stage using a Pearson correlation coefficient method;

[0058] For the feature-stage correlation matrix, a feature selection model is used to screen material features and update the feature weights and the feature priorities.

[0059] The processing stages are divided into roughing, finishing and polishing stages, and different stages have different sensitivities to material characteristics. The current processing stage can be determined by parameters such as current signal fluctuations, processing time and feed speed. As shown in Table 1, the pulse width, peak current and feed speed are used to match rules and quickly output the current processing stage.

[0060] Table 1 Processing stage rules

[0061]

[0062] Next, the current signals and vibration signals of different processed materials at different processing stages are collected to construct a data set. Features such as pulse density, current fundamental wave energy, harmonic energy ratio, and vibration signal main frequency band energy are extracted. The Pearson correlation coefficient is used to calculate the correlation between the eigenvalue and the processing stage, and the feature importance is obtained, which is expressed as:

[0063] ;

[0064] in, is the feature importance, For the The characteristic value of the data sample, For the The processing stages corresponding to each sample are set as follows: roughing = 1, finishing = 2, polishing = 3, is the mean value of this feature for all samples, Processing stage means for all samples.

[0065] According to the absolute value of feature importance, the features are sorted in descending order, and the feature priority is assigned according to the sorting result (the features with higher ranking have higher priority, and the features with lower ranking have lower priority). The optimal feature combination is selected according to the priority and applied to the judgment of the current processing stage to obtain the feature-stage correlation matrix, as shown in Table 2. The weight and priority of each feature corresponding to different processing stages can be used to screen the top J% of material features as input features for judging the material category.

[0066] Table 2 Feature-stage correlation matrix

[0067]

[0068] By optimizing the feature selection strategy and dynamically adjusting the features to avoid redundant calculations, the accuracy of processing material classification is effectively improved. When using the same data set, compared with fixed features, the material recognition accuracy is improved by 3.4%. At the same time, the processing stage discrimination speed is also improved by 23.7%, and the feature extraction reasoning speed is increased by about 1.9 times, which improves processing efficiency and system response speed.

[0069] Furthermore, the process of updating the feature weights includes:

[0070] ;

[0071] in, is the feature weight of feature i at time t+1, is the feature weight of the feature i at time t, is the change in classification accuracy of the feature i, is the change in the eigenvalue of the feature i, is the forgetting factor, which is set to 0.8 based on expert experience.

[0072] in, It indicates the change in classification accuracy when a feature is removed or added. For example, if the model accuracy decreases after removing feature i, the contribution of this feature is large and its weight should be increased. is the small change in the feature value, that is, the value of the feature in the new sample minus the value of the feature in the previous sample. If the change is large, it means that the feature may play a more important role in the new data and its weight needs to be adjusted dynamically.

[0073] Specifically, the feature selection model collects the classification results of the most recent K samples (whether the feature selection is correct), and calculates the contribution of each feature to the classification accuracy. If a feature can effectively reflect the processing status in recent samples, its weight should be increased. If a feature performs poorly recently and has a weak ability to distinguish the processing status, its weight should be reduced. The feature weights are updated to ensure smooth adjustment.

[0074] In addition, when the processing stage switches, such as from roughing to finishing, the feature selection is smoothly adjusted to avoid mutations that lead to classification errors. It can be expressed as:

[0075] ;

[0076] in, is the smoothed eigenvalue, is the eigenvalue of the previous stage, is the characteristic value of the current stage, is the smoothing factor, which is set to 0.8 and becomes 0 after a certain period of time.

[0077] During the processing, the feature selection model continuously optimizes the feature weights, effectively improving the system's adaptability to different processing conditions and enabling it to better adapt to the processing characteristics of different materials. In addition, when the processing stage switches, the system can smoothly adjust the feature selection strategy, effectively avoiding material classification errors caused by feature mutations, and ensuring the accuracy and continuity of material classification during the processing process.

[0078] Furthermore, the material current signal characteristics and the material vibration signal characteristics include:

[0079] The material current signal characteristics include: time domain characteristics, including waveform distortion rate, pulse density and peak-to-peak ratio; frequency domain characteristics, including fundamental wave energy ratio, third harmonic amplitude and band energy entropy; time-frequency domain characteristics, including wavelet energy moment and time-frequency ridge slope; statistical characteristics, including kurtosis coefficient and waveform factor;

[0080] The material vibration signal characteristics include: frequency domain analysis characteristics, including main frequency band energy ratio and harmonic wave attenuation rate; time-frequency analysis characteristics, including wavelet packet node energy and instantaneous frequency variance; nonlinear characteristics, including Lyapunov exponent and fractal dimension; statistical characteristics, including zero crossing rate and kurtosis skewness joint factor.

[0081] Specifically, in this embodiment, the characteristic parameters of the material current signal and the material vibration signal mainly come from the current sensor and the vibration sensor, including the Hall current sensor, the anti-aliasing filter and other key components. The material current signal is mainly used to analyze the characteristics of discharge stability, material conductivity, etc., while the vibration signal is used to evaluate the material hardness, processing stability and wire breakage risk. Through signal conditioning, data acquisition and digital signal processing, combined with multi-dimensional analysis of time domain, frequency domain, time-frequency domain and statistical characteristics, the key material characteristic information in the processing process can be accurately extracted, thereby improving the recognition accuracy and adaptability of various materials, and also improving the overall processing quality and efficiency.

[0082] As shown in Table 3, the calculation method and application of material current signal characteristics are given, where is the RMS current, is the average current, To set the ground discharge pulse threshold, is the counting function, is the unit time window, and The current signal The maximum and minimum values ​​of is the signal amplitude at 50Hz after FFT (Fast Fourier Transform) calculation, is the total energy of the entire spectrum, is the harmonic energy at 150Hz, is the energy proportion of the kth frequency band. In addition, the statistical features adopt conventional formulas to detect abnormal peaks and distinguish between cemented carbide and tool steel, respectively.

[0083] Table 3 Examples of material current signal characteristics

[0084]

[0085] In the material vibration signal analysis, the calculation method of wavelet packet node energy adopts 5-layer db6 wavelet packet decomposition. By extracting the energy value of the 5th point of the 3rd layer, it can be used to effectively identify the aluminum alloy sticking phenomenon. At the same time, the instantaneous frequency is obtained by Hilbert transform, and the variance within the 20ms window is calculated, which can monitor the jitter of the electrode wire in real time. The Lyapunov index calculates the maximum index value of the system after phase space reconstruction to warn of the possible chaotic wire breakage risk. The fractal dimension is calculated using the Katz algorithm to quantify the complexity of the vibration signal. In addition, parameters such as the main frequency band energy ratio, harmonic attenuation rate, zero crossing rate and kurtosis skewness joint factor are calculated using conventional formulas, which are used to associate material hardness, reflect system damping characteristics, distinguish hard / soft material processing status, and quantify vibration impact characteristics, thereby providing strong support for the comprehensive analysis of material vibration signals.

[0086] Further, Figure 2 The present invention provides a flow chart for identifying material categories, such as Figure 2 As shown in the figure, after identifying the processing stage, a stage-specific subtree set is set for each processing stage, including the rough processing stage subtree, the fine processing stage subtree, and the polishing stage subtree. Each stage subtree is trained for different processing modes. Since the classification accuracy of different processing stages is different, simple voting may lead to low-quality results. The dynamic weight voting mechanism is used to identify the voting weights of different processing stages, and the final classification is obtained after weighted fusion. The dynamic weight voting mechanism is expressed as:

[0087] ;

[0088] in, is the voting weight, is the stage classification accuracy, is the sum of the classification accuracy of all stages, is the number of samples processed by the subtree set at this stage, is the total number of samples of all stage subtree sets.

[0089] Furthermore, the model optimization unit includes:

[0090] If there is a new sample, check whether the new sample is misclassified; if misclassified, put the new sample into a buffer pool for periodically updating the stage-specific subtree set;

[0091] If the error rate of the stage-specific subtree set is greater than a preset threshold and the classification accuracy of the stage-specific subtree set decreases by more than M% for N consecutive times, the subtree is replaced.

[0092] Specifically, in this embodiment, the classification of new samples is monitored. If the classification is wrong, it is stored in the error sample buffer pool, and the model update is triggered every 50 groups of samples. By adopting a dynamic subtree optimization mechanism, combined with error sample buffering, error rate monitoring and classification accuracy change analysis, adaptive model optimization is achieved: when the error rate exceeds 15%, the full replacement strategy is enabled, and new subtrees need to be retrained, and the old subtrees are directly replaced with newly trained subtrees. The error sample buffer pool is cleared and accumulation starts again to ensure the reliability of the model under extreme degradation conditions; when the classification accuracy drops by more than 5% for three consecutive rounds, a partial replacement strategy is adopted to replace only the worst 30% subtrees, and new subtrees are trained to fill the deleted subtrees, so that the model is smoothly optimized to avoid short-term fluctuations affecting the overall performance. Compared with the traditional random forest model, the error rate of this method is reduced by 12% to 15%, and the classification accuracy is improved by 4.3% to 5.6%, which enhances the long-term stability of the model and its adaptability to multiple materials. In addition, only some subtrees are replaced at critical moments, which reduces unnecessary computing burden and model update calculation amount, effectively improves optimization efficiency, ensures material identification accuracy, and thus improves the efficiency and accuracy of wire cutting machines in multi-variety and small-batch mold processing.

[0093] Dual-loop adaptive control module, such as Figure 3 As shown, it includes: an outer loop control unit, which is used to retrieve basic process parameters for the processing material and perform initial compensation according to historical processing records; an inner loop control unit, which is used to adjust real-time fine-tuning parameters using a fuzzy PID model according to dual-target error values;

[0094] Among them, after identifying the type of material currently being processed, the recommended basic process parameters, including pulse width, peak current, and wire speed, are queried from the database, as shown in Table 4. According to the conductivity of the material, such as graphite, which has good conductivity, it is set to a low voltage. If there is a historical processing record for the material, its past processing error data is retrieved. If the historical error exceeds the threshold (such as ±10%), the basic process parameters are adaptively adjusted to compensate for interference such as environmental fluctuations and equipment loss. For example, the difference between the current working fluid temperature and the standard temperature is multiplied by the temperature compensation coefficient, which is the compensation item for the basic process parameters.

[0095] Table 4 Basic process parameter examples

[0096]

[0097] Furthermore, the adjustment process of the real-time fine-tuning parameters includes:

[0098] Calculate the surface roughness deviation and material removal rate, expressed as:

[0099] ;

[0100] in, is the surface roughness deviation, is the current processing surface roughness (measured by roughness sensor), is the target value of machining surface roughness (material process parameter database), is the material removal rate, is the feed speed, is the cutting cross-sectional area, For processing time.

[0101] For each of the processing stages, the processing weights of the surface roughness deviation and the material removal rate are allocated, and the dual-objective error value is obtained using a weighted summation method;

[0102] If the dual-objective error value is less than the optimization target value, the real-time fine-tuning parameter is adjusted using the fuzzy PID model; otherwise, the real-time fine-tuning parameter is maintained.

[0103] Among them, the real-time fine-tuning parameters include the dynamic correction amount of the feed speed.

[0104] Specifically, in this embodiment, in the rough machining stage, The processing weight is 0.3, The processing weight is 0.7 to prioritize efficiency; in the finishing stage, The processing weight is 0.8, The processing weight is 0.2 to give priority to surface quality; in the finishing stage, The processing weight is 0.9, The processing weight is 0.1, and only fine-tuning is performed to maintain accuracy. The dual-objective error value is calculated according to the processing weight. If the dual-objective error value is less than the optimization target value 0.1, the fuzzy PID model is used to adjust the real-time fine-tuning parameters; otherwise, the real-time fine-tuning parameters are maintained.

[0105] In the outer control unit, the system dynamically adjusts the basic process parameters based on the material database and historical error analysis, effectively reducing the processing errors caused by material changes. The inner control unit uses fuzzy PID control to adjust the dynamic correction amount of the feed speed in real time during the processing, and sets the corresponding weight according to the different stages of processing to accurately compensate for the processing errors, thereby improving the efficiency and accuracy of the wire cutting machine in the processing of small batches of molds of various varieties.

[0106] Furthermore, the fuzzy PID model includes:

[0107] Calculate the change rate of the dual target error values ​​to obtain the material error change rate;

[0108] The material error change rate and the dual-target error value are set as material input variables, and the material proportional coefficient adjustment amount, the material integral coefficient adjustment amount and the material differential coefficient adjustment amount are set as parameter output variables;

[0109] Among them, the material proportional coefficient adjustment amount is to adjust the PID proportional gain, the material integral coefficient adjustment amount is to adjust the PID integral gain, and the material differential coefficient adjustment amount is to adjust the PID differential gain.

[0110] Fuzzifying the material input variable and the parameter output variable to obtain a material fuzzy input variable and a material fuzzy output variable;

[0111] Setting a material fuzzy rule table, for obtaining the material fuzzy output variable according to the material fuzzy input variable;

[0112] The fuzzy output variable is defuzzified using the center of gravity method, and the PID parameters are adjusted to obtain the real-time fine-tuning parameters.

[0113] Specifically, in order to enable the computer to process continuous error values, fuzzification is required. The fuzzification processing of material input variables and parameter output variables is shown in Tables 5 and 6. For example, when the dual-objective error value is small (S), it means that the error is small and the target is basically achieved. When the dual-objective error value is medium (M), the error is within an acceptable range, but there is room for optimization.

[0114] Table 5 Material input variables

[0115]

[0116] Table 6 Parameter output variables

[0117]

[0118] Combining dual target errors and error rate of change , determines how to adjust the PID parameters, as shown in Table 7. An example of fuzzy rule representation is given based on expert experience. For example, if the error is very small (S) and the error is increasing rapidly (PB), the material proportional coefficient adjustment amount is kept unchanged, and the material integral coefficient adjustment amount and the material differential coefficient adjustment amount are reduced to prevent overshoot.

[0119] In addition, in order to prevent sudden changes in parameters, limit adjustment is performed, which is expressed as:

[0120] ;

[0121] in, is the original setting value, is the adjustment amount of fuzzy PID output, is the maximum allowed value and can be set to , is the minimum allowed value and can be set to .

[0122] Table 7 Fuzzy rule table example

[0123]

[0124] By adopting the fuzzy PID control method, with the help of dual-target error feedback and fuzzy logic reasoning, the PID parameters are adjusted in real time. Compared with the traditional PID control method, the control accuracy is improved by 7.8%, thereby improving the efficiency and accuracy of the wire cutting machine in the processing of multi-variety and small-batch molds.

[0125] The energy consumption management optimization module is used to identify non-processing periods and processing periods according to trigger conditions and optimize energy consumption using dynamic control strategies.

[0126] The present invention realizes high-precision material identification through a multimodal signal material identification module, integrating current and vibration signals, and combining a dynamic weight voting mechanism. For different modes such as rough machining, fine machining and polishing, the stage-specific subtree set is used to optimize the recognition accuracy. At the same time, the system has real-time online learning capabilities to adapt to new materials and new processes, avoiding the limitations of traditional methods that rely on fixed process libraries. The outer loop control unit of the dual-loop adaptive control module automatically compensates for basic process parameters based on historical records to reduce errors and improve consistency; the inner loop control unit adopts fuzzy PID dynamic adjustment based on dual-objective optimization to ensure processing quality. The energy consumption management optimization module adopts energy-saving strategies during non-processing periods to reduce standby energy consumption; during processing periods, energy efficiency is optimized to reduce ineffective energy consumption. The present invention effectively improves the efficiency and accuracy of wire cutting machines in multi-variety small-batch mold processing.

[0127] Embodiment 2:

[0128] In order to verify the actual application effect of the first embodiment, a wire cutting processing workshop of a precision manufacturing company is used as a test environment. It is aimed at small batch production of multiple varieties and frequent replacement of processing materials, such as SKD11 high-hardness steel, 6061 aluminum alloy, graphite EDM-3, etc., and has high requirements for processing accuracy. By deploying a wire cutting machine feed speed control system, a wire cutting machine feed speed control method, such as Figure 4 The specific implementation process is as follows:

[0129] According to the adaptive feature selection mechanism, the material current signal features and material vibration signal features are extracted;

[0130] An improved random forest model is established based on the material current signal characteristics and the material vibration signal characteristics. The random forest model sets stage-specific subtree sets for different processing stages, and uses a dynamic weight voting mechanism and a weighted fusion method to identify processing materials; wherein, the number of trees in the rough processing subtree set is 30, and the maximum depth is 8, the number of trees in the fine processing subtree set is 50, and the maximum depth is 12, and the number of trees in the polishing subtree set is 20, and the maximum depth is 6.

[0131] Performing online learning and model updating on the random forest model;

[0132] Among them, in this embodiment, the average accuracy of the traditional random forest model is 90.95%, while the average accuracy of the random forest model improved by the present invention reaches 94.63%, especially in the recognition of graphite EDM-3, the accuracy is increased from 87.1% to 92.4%, which significantly improves the overall performance.

[0133] Retrieving basic process parameters for the processing material and performing initial compensation according to historical processing records;

[0134] According to the dual-objective error value, the real-time fine-tuning parameters are adjusted using the fuzzy PID model;

[0135] Identify non-processing and processing periods based on trigger conditions and use dynamic control strategies to optimize energy consumption.

[0136] Among them, in this embodiment, there are 4 trigger conditions. Trigger condition 1 is the discharge signal state. The sliding window counting method is adopted. If there is no discharge signal for more than 5 minutes, the non-processing period is entered. If the discharge signal is restored, the timing is reset, and monitoring continues to enter the processing period; trigger condition 2 is the servo motor movement state. If the XYZ axis is stationary for more than 10 seconds, the non-processing period is entered. If the axis movement is restored, the processing period is entered; trigger condition 3 is the process task queue. If the task queue is empty, the non-processing period is entered. If the task queue is not empty, the non-processing period is entered; trigger condition 4 is the environmental sensor signal. If there is no operation in the workshop for more than 5 minutes, the non-processing period is entered.

[0137] Based on the triggering conditions, this embodiment adopts different dynamic control strategies in the non-processing period and the processing period. The dynamic control strategies specifically include:

[0138] Energy-saving strategy during non-processing periods: the working fluid circulation system is reduced to the lowest flow mode (1L / min), the servo drive system uses the zero torque holding mode, the pulse power supply turns off the filtering and pre-charging circuits, and the workshop lighting is intelligently dimmed by induction;

[0139] Energy efficiency optimization strategy during processing period, working fluid flow optimization is adaptively adjusted according to the inter-electrode status. For example, using rule-based control method, when the inter-electrode voltage fluctuation is greater than 10V, the flow rate is increased to 4L / min, and when the inter-electrode voltage fluctuation is less than 2V, the flow rate is reduced to 1.5L / min; pulse power supply gap voltage regulation, the discharge gap voltage is adjusted from 120V to 90V.

[0140] Under the same workshop environment, the method of the present invention using trigger conditions and dynamic energy consumption optimization strategy is used as Scheme 1, and the method of the present invention not using trigger conditions and dynamic energy consumption optimization strategy is used as Scheme 2. Scheme 1 is compared with Scheme 2. The standby energy consumption of the wire cutting machine is reduced by 24%, reducing unnecessary power loss. The energy saving during the processing period is 15%, the working fluid discharge energy consumption is optimized, the production cost is reduced, and the equipment operation efficiency is improved.

[0141] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A wire cutting machine feed speed control system, characterized in that: include: A multimodal signal material recognition module includes: a signal feature extraction unit, which is used to extract material current signal features and material vibration signal features according to an adaptive feature selection mechanism; a material classification unit, which is used to establish an improved random forest model based on the signal feature extraction unit, wherein the random forest model sets stage-specific subtree sets for different processing stages, and uses a dynamic weight voting mechanism and a weighted fusion method to identify processing materials; a model optimization unit, which is used to perform online learning and model update on the random forest model; the adaptive feature selection mechanism includes: setting preset rules for determining the processing stage of the wire cutting machine; establishing a feature-stage correlation matrix, which is used to set feature weights and feature priorities for the material current signal features and the material vibration signal features according to the processing stage using the Pearson correlation coefficient method; for the feature-stage correlation matrix, using a feature selection model to screen material features and update the feature weights and feature priorities; A dual-loop adaptive control module, comprising: an outer-loop control unit, used to retrieve basic process parameters for the processing material and perform initial compensation according to historical processing records; an inner-loop control unit, used to adjust real-time fine-tuning parameters using a fuzzy PID model according to a dual-objective error value; the adjustment process of the real-time fine-tuning parameters comprises: calculating surface roughness deviation and material removal rate; for each processing stage, allocating processing weights of the surface roughness deviation and the material removal rate, and obtaining the dual-objective error value using a weighted summation method; if the dual-objective error value is less than an optimization target value, adjusting the real-time fine-tuning parameters using the fuzzy PID model; otherwise, maintaining the real-time fine-tuning parameters; The fuzzy PID model includes: calculating the change rate of the dual-objective error value to obtain the material error change rate; setting the material error change rate and the dual-objective error value as material input variables, and setting the material proportional coefficient adjustment amount, the material integral coefficient adjustment amount and the material differential coefficient adjustment amount as parameter output variables; fuzzifying the material input variable and the parameter output variable to obtain a material fuzzy input variable and a material fuzzy output variable; setting a material fuzzy rule table for obtaining the material fuzzy output variable according to the material fuzzy input variable; using the centroid method to defuzzify the fuzzy output variable, and adjusting the PID parameters to obtain the real-time fine-tuning parameters; The energy consumption management optimization module is used to identify non-processing periods and processing periods according to trigger conditions and optimize energy consumption using dynamic control strategies.

2. A wire cutting machine feed speed control system according to claim 1, characterized in that: The process of updating the feature weights includes: ； in, is the feature weight of feature i at time t+1, is the feature weight of the feature i at time t, is the change in classification accuracy of the feature i, is the change in the eigenvalue of the feature i, For the forgetting factor.

3. A wire cutting machine feed speed control system according to claim 1, characterized in that: The material current signal characteristics and the material vibration signal characteristics include: The material current signal characteristics include: time domain characteristics, including waveform distortion rate, pulse density and peak-to-peak ratio; frequency domain characteristics, including fundamental wave energy ratio, third harmonic amplitude and band energy entropy; time-frequency domain characteristics, including wavelet energy moment and time-frequency ridge slope; statistical characteristics, including kurtosis coefficient and waveform factor; The material vibration signal characteristics include: frequency domain analysis characteristics, including main frequency band energy ratio and harmonic wave attenuation rate; time-frequency analysis characteristics, including wavelet packet node energy and instantaneous frequency variance; nonlinear characteristics, including Lyapunov exponent and fractal dimension; statistical characteristics, including zero crossing rate and kurtosis skewness joint factor.

4. A wire cutting machine feed speed control system according to claim 1, characterized in that: The model optimization unit comprises: If there is a new sample, check whether the new sample is misclassified; if misclassified, put the new sample into a buffer pool for periodically updating the stage-specific subtree set; If the error rate of the stage-specific subtree set is greater than a preset threshold and the classification accuracy of the stage-specific subtree set decreases by more than M% for N consecutive times, the subtree is replaced.

5. A wire cutting machine feed speed control method, characterized in that: include: According to the adaptive feature selection mechanism, the material current signal features and material vibration signal features are extracted; According to the material current signal characteristics and the material vibration signal characteristics, an improved random forest model is established, wherein the random forest model sets stage-specific subtree sets for different processing stages, and uses a dynamic weight voting mechanism and a weighted fusion method to identify processing materials; The random forest model is subjected to online learning and model updating; the adaptive feature selection mechanism includes: setting preset rules for determining the processing stage of the wire cutting machine; establishing a feature-stage correlation matrix for setting feature weights and feature priorities for the material current signal features and the material vibration signal features according to the processing stage using the Pearson correlation coefficient method; for the feature-stage correlation matrix, using a feature selection model to screen material features and update the feature weights and feature priorities; Retrieving basic process parameters for the processing material and performing initial compensation according to historical processing records; According to the dual-objective error value, the real-time fine-tuning parameter is adjusted using the fuzzy PID model; the adjustment process of the real-time fine-tuning parameter includes: calculating the surface roughness deviation and the material removal rate; for each of the processing stages, the processing weights of the surface roughness deviation and the material removal rate are assigned, and the dual-objective error value is obtained using the weighted summation method; if the dual-objective error value is less than the optimization target value, the real-time fine-tuning parameter is adjusted using the fuzzy PID model; otherwise, the real-time fine-tuning parameter is maintained; The fuzzy PID model includes: calculating the change rate of the dual-objective error value to obtain the material error change rate; setting the material error change rate and the dual-objective error value as material input variables, and setting the material proportional coefficient adjustment amount, the material integral coefficient adjustment amount and the material differential coefficient adjustment amount as parameter output variables; fuzzifying the material input variable and the parameter output variable to obtain a material fuzzy input variable and a material fuzzy output variable; setting a material fuzzy rule table for obtaining the material fuzzy output variable according to the material fuzzy input variable; using the centroid method to defuzzify the fuzzy output variable, and adjusting the PID parameters to obtain the real-time fine-tuning parameters; Identify non-processing and processing periods based on trigger conditions and use dynamic control strategies to optimize energy consumption.

Images