Force pre-warning method and device for cutting force measuring tool holder, and electronic equipment

Abstract

This invention relates to the field of cutting technology, providing a force measurement and early warning method, device, and electronic equipment for a cutting force measuring tool holder. The method includes: constructing a digital twin model of the cutting process based on first preset parameters to describe the dynamic behavior of cutting force under normal cutting conditions as cutting conditions change; using normal cutting force feature data obtained from the digital twin model as training samples to train a normal cutting state model; inputting real-time cutting force feature data extracted from real-time cutting force data into the normal cutting state model, and calculating the difference between the real-time cutting force feature information output by the normal cutting state model and the real-time cutting force feature data; determining cutting state anomalies based on the difference and an early warning threshold, and performing early warning processing; the early warning threshold is dynamically determined. This invention achieves accurate early warning of cutting anomalies under varying working conditions based on digital twins and machine learning.

Description

Technical Field

[0001] This invention relates to the field of cutting technology, and in particular to a force measurement and early warning method and device for a cutting force measuring tool holder, as well as electronic equipment. Background Technology

[0002] In CNC machining, cutting force is a crucial parameter reflecting the tool load state and machining stability. With the development of cutting force measurement tool holder technology, real-time acquisition of cutting force has been gradually realized. In existing technologies, force measurement early warning methods mainly rely on fixed thresholds or empirical rules for judgment. That is, the operator sets an upper limit value for the cutting force based on experience, and an alarm is triggered when the real-time acquired cutting force exceeds this threshold.

[0003] However, the above methods have the following shortcomings: the warning threshold relies on human experience, making it difficult to adapt to changes in different cutting parameters and workpiece materials; they cannot characterize the dynamic behavior of cutting force changes with time and working conditions; and they are prone to false alarms or missed alarms under complex machining conditions. For example, when cutting parameters change significantly, the normal cutting force itself may exceed the fixed threshold, leading to a false alarm, while when the tool wears slowly, the gradually increasing cutting force may cause chipping before exceeding the threshold, resulting in a missed alarm. Therefore, how to achieve accurate early warning of cutting anomalies under varying working conditions without frequent manual adjustment of the threshold is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0004] This invention provides a force measurement early warning method, device, and electronic equipment for cutting force measuring tool holders, which solves the problems of fixed threshold early warning in the prior art, which is difficult to adapt to changes in working conditions and has a high false alarm and missed alarm rate. Based on digital twins and machine learning, it realizes accurate early warning of cutting abnormalities under changing working conditions.

[0005] In a first aspect, the present invention provides a force measurement early warning method for a cutting force-measuring tool holder, comprising: A digital twin model of the cutting process is constructed based on a first preset parameter; wherein, the digital twin model is used to describe the dynamic behavior of the cutting force under normal cutting conditions as the cutting condition parameters change, and the first preset parameter includes tool structure parameters, the cutting condition parameters, and measured cutting force data under historical normal cutting conditions; Under normal cutting conditions, the normal cutting force feature data obtained by the digital twin model is used as training samples to train a normal cutting state model; wherein, the normal cutting state model is used to reconstruct or fit the normal cutting force feature data. The real-time cutting force feature data extracted from the real-time cutting force data is input into the normal cutting state model, and the difference between the real-time cutting force feature information output by the normal cutting state model and the real-time cutting force feature data is calculated. An abnormal cutting state is determined based on the difference and the warning threshold, and a warning is issued; wherein, the warning threshold is dynamically determined based on the theoretical normal range predicted by the digital twin model for the normal cutting force behavior under the current cutting condition.

[0006] In some embodiments, constructing a digital twin model of the cutting process based on a first preset parameter includes: Based on the tool structure parameters and the cutting condition parameters, an initial digital twin model is constructed using the cutting force mechanism formula; The model parameters of the initial digital twin model are calibrated using the measured cutting force data under the historical normal cutting conditions, so that the error between the theoretical cutting force output by the digital twin model and the actual normal cutting force is less than or equal to a preset error.

[0007] In some embodiments, obtaining normal cutting force characteristic data using the digital twin model includes: The candidate cutting force features are mapped to the virtual state space corresponding to the digital twin model; When the candidate cutting force feature falls within the theoretical normal range, the corresponding candidate cutting force feature is marked as a normal sample and used to train the normal cutting state model; wherein, the candidate cutting force feature comes from the measured cutting force data under historical normal cutting conditions or the generated data of the digital twin model.

[0008] In some embodiments, dynamically determining the warning threshold includes: The current cutting condition parameters are input into the digital twin model to obtain the theoretical normal range of the normal cutting force characteristics; wherein, the theoretical normal range varies with the change of a second preset parameter, the second preset parameter including at least one of depth of cut, feed rate or spindle speed; The warning threshold is determined based on the theoretical normal range.

[0009] In some embodiments, extracting real-time cutting force characteristic data from real-time cutting force data includes: The real-time cutting force data is filtered and denoised, and at least one real-time cutting force feature data is extracted from the preprocessed real-time cutting force data, including average cutting force, peak cutting force, cutting force fluctuation amplitude, cutting force change rate, and multi-directional cutting force synthesis feature.

[0010] In some embodiments, after determining the abnormal cutting state based on the difference degree and the warning threshold, and performing warning processing, the method further includes: The real-time cutting force characteristic data and cutting condition parameters corresponding to the early warning processing are fed back to the digital twin model and the normal cutting state model to update the preset parameters of the digital twin model and the decision boundary of the normal cutting state model.

[0011] In some embodiments, the normal cutting state model is an autoencoder, and the difference is the reconstruction error; or, the normal cutting state model is a single-class support vector machine, and the difference is a distance metric in the feature space.

[0012] In some embodiments, determining an abnormal cutting state based on the difference degree and the warning threshold, and performing warning processing, includes: When the degree of difference is greater than or equal to the warning threshold, the ratio of the degree of difference to the warning threshold is output as the warning confidence level; and / or, the main feature type causing the anomaly is output according to the degree of contribution of each feature component in the real-time cutting force feature data to the degree of difference.

[0013] Secondly, the present invention also provides a force measurement early warning device for a cutting force measuring tool holder, comprising: The twin model construction module is used to construct a digital twin model of the cutting process based on a first preset parameter; wherein, the digital twin model is used to describe the dynamic behavior of the cutting force under normal cutting conditions as the cutting condition parameters change, and the first preset parameter includes tool structure parameters, the cutting condition parameters, and measured cutting force data under historical normal cutting conditions; The machine learning training module is used to train a normal cutting state model under normal cutting conditions by using the normal cutting force feature data obtained from the digital twin model as training samples; wherein, the normal cutting state model is used to reconstruct or fit the normal cutting force feature data. The difference parameter calculation module is used to input the real-time cutting force feature data extracted from the real-time cutting force data into the normal cutting state model, and calculate the difference between the real-time cutting force feature information output by the normal cutting state model and the real-time cutting force feature data. The dynamic early warning processing module is used to determine the abnormal cutting state based on the difference degree and the early warning threshold, and to perform early warning processing; wherein, the early warning threshold is dynamically determined based on the theoretical normal range of the normal cutting force behavior predicted by the digital twin model under the current cutting conditions.

[0014] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to execute the force measurement early warning method for the cutting force measuring tool holder as described in the first aspect through the computer program.

[0015] This invention uses a normal cutting state model to characterize the behavior pattern of cutting forces. The difference degree calculation comprehensively reflects the deviation of cutting forces across multiple feature dimensions, thus its ability to identify complex anomalies such as slow tool wear and cutting force waveform distortion is significantly superior to fixed threshold methods. Furthermore, the warning threshold is not fixed but dynamically determined based on the prediction results of the digital twin model for the current cutting conditions, thereby avoiding false alarms or missed alarms caused by changes in operating conditions. In addition, it does not require changes to existing hardware configurations and is easy to deploy on existing CNC equipment. Since the normal cutting state model learns the statistical regularities of normal patterns, even minor deviations from the normal pattern can be promptly captured through an increase in the difference degree, thereby achieving early warning and reducing the risks of tool overload, chipping, and machining instability. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a force measurement early warning method for a cutting force measuring tool holder provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the specific process of a force measurement early warning method for a cutting force measuring tool holder provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a force measuring and early warning device for a cutting force measuring tool holder provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0019] In CNC machining, cutting force is a crucial parameter reflecting the tool load state and machining stability. With the development of cutting force measurement tool holder technology, real-time acquisition of cutting force has been gradually realized. However, existing force measurement early warning methods still mainly rely on fixed thresholds or empirical rules for judgment, which has the following shortcomings: the early warning threshold depends on human experience, making it difficult to adapt to different cutting parameters and materials; it cannot characterize the dynamic behavior of cutting force changes with time and working conditions; and it is prone to false alarms or missed alarms under complex machining conditions.

[0020] To address the aforementioned technical problems, this invention proposes a cutting force measuring tool holder force measurement early warning method based on digital twins and machine learning. By performing digital twin modeling on real-time cutting force data collected by the cutting force measuring tool holder, and utilizing machine learning algorithms to learn the cutting force behavior characteristics under normal cutting conditions, intelligent identification and early warning of abnormal cutting conditions are achieved, thereby reducing the risks of tool overload, chipping, and machining instability. Therefore, this invention enables the cutting force measuring tool holder to accurately identify abnormal cutting conditions under complex machining conditions. Specifically, it utilizes a digital twin model to characterize normal cutting force behavior, improving the accuracy of early warning judgment; it achieves automatic identification of abnormal conditions through machine learning algorithms, reducing manual intervention; the early warning threshold has adaptive capabilities to adapt to different working conditions; and it requires no additional hardware, making it easy to implement in engineering.

[0021] Figure 1 This is a flowchart illustrating a force measurement early warning method for a cutting force measuring tool holder according to an embodiment of the present invention. The force measurement early warning method for the cutting force measuring tool holder can be executed by the force measurement early warning device for the cutting force measuring tool holder provided in this embodiment of the invention. This force measurement early warning device can be implemented using software and / or hardware. For example... Figure 1 As shown, the force measurement early warning method for cutting force measuring tool holders includes the following steps: S101. Construct a digital twin model of the cutting process based on the first preset parameters; wherein, the digital twin model is used to describe the dynamic behavior of the cutting force under normal cutting conditions as the cutting condition parameters change, and the first preset parameters include tool structure parameters, cutting condition parameters, and measured cutting force data under historical normal cutting conditions.

[0022] S102. Under normal cutting conditions, the normal cutting force feature data obtained by the digital twin model is used as training samples to train a normal cutting state model; wherein, the normal cutting state model is used to reconstruct or fit the normal cutting force feature data.

[0023] S103. Input the real-time cutting force characteristic data extracted from the real-time cutting force data into the normal cutting state model, and calculate the difference between the real-time cutting force characteristic information output by the normal cutting state model and the real-time cutting force characteristic data.

[0024] S104. Determine the abnormal cutting state based on the degree of difference and the warning threshold, and perform warning processing; wherein, the warning threshold is dynamically determined based on the theoretical normal range predicted by the digital twin model for the normal cutting force behavior under the current cutting conditions.

[0025] Specifically, a cutting force-measuring toolholder refers to a toolholder device with a built-in force sensor, installed between the spindle and the tool of a CNC machine tool, capable of measuring and outputting multi-dimensional cutting force data on the tool in real time during the cutting process. A digital twin model refers to a mathematical model constructed in virtual space that corresponds one-to-one with the actual cutting process. This model, based on physical mechanisms and historical data, can simulate and predict the dynamic behavior of cutting force under normal cutting conditions as cutting parameters change. The first preset parameters refer to a set of input parameters used to construct the digital twin model, specifically including tool structure parameters, cutting condition parameters, and measured cutting force data under historical normal cutting conditions. Tool structure parameters refer to parameters describing the geometric characteristics of the tool and the mechanical properties of the toolholder, including but not limited to tool diameter, number of teeth, overhang length, helix angle, and rake angle. Cutting condition parameters refer to a set of parameters describing the machining conditions, including but not limited to depth of cut, feed rate, spindle speed, width of cut, and workpiece material properties. The measured cutting force data under normal historical cutting conditions refers to the sequence of cutting force values ​​actually collected and recorded by the cutting force measuring tool holder during historical machining processes when the machining state is confirmed to be normal, that is, the tool has no abnormal wear, the workpiece material is uniform, and the cutting process is stable.

[0026] Normal cutting conditions refer to a machining state where the tool is in good condition, cutting parameters are reasonable, there is no abnormal vibration or chipping, and the machining process is stable. Normal cutting force characteristic data refers to the set of characteristic parameters extracted from cutting force data under normal cutting conditions that characterize the behavior pattern of normal cutting forces, such as average cutting force, peak cutting force, and cutting force fluctuation amplitude. Machine learning algorithms refer to computer algorithms that can automatically learn patterns and rules from data, used to learn the inherent distribution laws of normal cutting force characteristic data. A normal cutting state model refers to a mathematical model trained using machine learning algorithms, which can reconstruct or fit the input normal cutting force characteristic data, i.e., reproduce the expected form of normal cutting force data.

[0027] Real-time cutting force data refers to the raw, unprocessed or pre-processed cutting force values ​​collected and output in real time by the cutting force measuring tool holder during the current machining process. Real-time cutting force feature data refers to the set of feature parameters obtained from the real-time cutting force data after preprocessing and feature extraction, used for judging the current state; its dimensions are consistent with normal cutting force feature data. Real-time cutting force feature information refers to the result output by the normal cutting state model after processing the input real-time cutting force feature data. Difference degree is a quantitative index used to measure the degree of difference between the real-time cutting force feature information output by the normal cutting state model and the input real-time cutting force feature data. The greater the difference degree, the greater the deviation between the current cutting state and the normal cutting state. Warning threshold refers to a dynamically determined numerical limit used to determine whether the difference degree reaches a level requiring an alarm. When the difference degree exceeds this threshold, an abnormal cutting risk is determined. Theoretical normal range refers to the numerical range or distribution area that the cutting force features should fall into under normal cutting conditions, predicted by the digital twin model based on the current cutting condition parameters. Early warning processing refers to a series of response actions performed when the difference exceeds the early warning threshold, including but not limited to issuing audible and visual alarm signals, displaying early warning information on the operation interface, and recording abnormal data.

[0028] First, a digital twin model is constructed, that is, a digital twin model of the cutting process is built based on a first set of preset parameters. Specifically, the system acquires tool structural parameters, such as tool diameter, number of teeth, and overhang length, as well as current or expected cutting condition parameters, such as depth of cut, feed rate, and spindle speed, and historical measured cutting force data under normal cutting conditions. Using these parameters, an initial mathematical model is established through cutting force mechanism formulas, such as the average force model or instantaneous rigid force model in the milling process. Then, the coefficients of this model are calibrated and corrected using historical measured data. The final digital twin model can describe how the cutting force will change over time under normal cutting conditions, given a set of cutting condition parameters, and what its average value and fluctuation range should be. This model is essentially a health benchmark model, representing the mechanical behavior that should exist under normal machining.

[0029] Secondly, the normal cutting state model is trained. Assuming the machine tool is under normal cutting conditions, the system uses the constructed digital twin model to acquire normal cutting force characteristic data. This acquisition can be achieved by selecting samples from historical measured data under normal cutting conditions that match the prediction range of the digital twin model, or by directly generating theoretical normal cutting force characteristic data from the digital twin model. These normal cutting force characteristic data are then used as training samples and input into a machine learning algorithm, such as an autoencoder or a single-class support vector machine, for training. After training, a normal cutting state model is obtained. The core function of this model is to reconstruct or fit the input cutting force characteristic data. That is, when a normal cutting force characteristic data is input into the normal cutting state model, the model can output a reconstruction result that is very close to the input; the normal cutting state model essentially learns the distribution pattern and intrinsic structure of the normal cutting force data in the feature space.

[0030] Then, the real-time difference is calculated. During actual machining, the cutting force measuring tool holder continuously collects real-time cutting force data. This real-time data is preprocessed and feature extracted to obtain real-time cutting force feature data. This real-time cutting force feature data is then input into a pre-trained normal cutting state model. The normal cutting state model outputs a reconstruction or fitting result, i.e., the real-time cutting force feature information, based on the normal pattern it has learned. Subsequently, the difference between the real-time cutting force feature information output by the normal cutting state model and the input real-time cutting force feature data is calculated. The physical meaning of the difference is the degree of deviation between the currently measured cutting force feature and the normal pattern described by the normal cutting state model. If the current machining state is normal, the difference should be small; if anomalies occur, such as tool wear, cutting overload, or hard material, the measured features will deviate from the normal pattern, and the difference will increase significantly.

[0031] Finally, warning judgment and processing are performed. Based on the theoretical normal range predicted by the digital twin model for normal cutting force behavior under the current cutting conditions, a warning threshold is dynamically determined. This theoretical normal range changes with the current cutting parameters, such as depth of cut and feed rate; therefore, the warning threshold is also dynamic and can adapt to different machining conditions. The calculated difference is compared with this dynamic warning threshold. When the difference exceeds the warning threshold, it is determined that there is an abnormal risk in the current cutting state, and warning processing is immediately carried out, such as issuing an alarm signal, prompting the operator on the display screen, and recording the abnormality log.

[0032] Therefore, this embodiment of the invention uses a normal cutting state model to characterize the behavior pattern of cutting forces. The difference degree calculation can comprehensively reflect the deviation of cutting forces in multiple feature dimensions. Thus, its ability to identify complex abnormalities such as slow tool wear and cutting force waveform distortion is significantly better than that of the fixed threshold method. Furthermore, the warning threshold is not fixed but dynamically determined based on the prediction results of the digital twin model for the current cutting conditions. When the cutting depth, feed rate, or spindle speed changes, the theoretical normal range is automatically adjusted, and the warning threshold changes accordingly, thereby avoiding false alarms or missed alarms caused by changes in operating conditions. In addition, it utilizes only the existing data acquisition capabilities and computing resources of the cutting force measuring tool holder, implemented through software algorithms, without changing the existing hardware configuration, making it easy to promote and deploy on existing CNC equipment. Moreover, since the normal cutting state model learns the statistical regularities of normal patterns, even small changes deviating from the normal pattern can be captured in a timely manner through an increase in the difference degree, thereby achieving early warning and reducing the risks of tool overload, chipping, and machining instability.

[0033] In some embodiments, constructing a digital twin model of the cutting process based on a first preset parameter includes: constructing an initial digital twin model using a cutting force mechanism formula based on tool structure parameters and cutting condition parameters; calibrating the model parameters of the initial digital twin model using measured cutting force data under historical normal cutting conditions, so that the error between the theoretical cutting force output by the digital twin model and the actual normal cutting force is less than or equal to a preset error.

[0034] Specifically, the cutting force mechanism formula refers to a mathematical expression based on metal cutting theory and mechanical principles, used to calculate the relationship between cutting force and cutting parameters. For example, in milling, the cutting force can be expressed as an empirical formula of the cutting force coefficient multiplied by the cutting area. The initial digital twin model refers to a digital twin model directly constructed using the cutting force mechanism formula based solely on tool structure parameters and cutting condition parameters, before historical data calibration. The parameters of the initial digital twin model are theoretical or empirical values ​​and have not yet been aligned with actual machining data. Calibration refers to the process of using measured cutting force data under historical normal cutting conditions, and adjusting the unknown coefficients or correction factors in the initial digital twin model through parameter identification methods such as least squares method and Kalman filtering, to reduce the error between the theoretical cutting force output by the initial digital twin model and the actual normal cutting force to an acceptable range. The preset error refers to a pre-set error tolerance value used to determine whether the calibration process meets the requirements. When the error between the theoretical cutting force and the actual normal cutting force is less than or equal to this preset error, the digital twin model can accurately reflect the normal cutting state.

[0035] First, obtain the tool structure parameters, such as tool diameter, number of teeth, helix angle, and overhang length, as well as cutting condition parameters, such as depth of cut, feed rate, spindle speed, and workpiece material. Based on these parameters, an initial digital twin model is established using the cutting force mechanism formula. The initial digital twin model can take the form of, for example, the theoretical cutting force equal to the product of the specific cutting force coefficient, depth of cut, feed per tooth, width of cut coefficient, and tool structure correction factor. The specific cutting force coefficient, correction factor, and other parameters use theoretical or empirical values ​​in the initial model.

[0036] Next, measured cutting force data under historical normal cutting conditions are acquired. This data is collected by the cutting force measuring tool holder during the historical process confirming normal and stable machining conditions. These measured data are compared with the theoretical cutting force output by the initial digital twin model under the same working conditions. Then, parameter identification algorithms, such as the least squares method, are used to adjust the adjustable parameters in the initial digital twin model, such as the specific cutting force coefficient, friction coefficient, and stiffness correction coefficient, so that the error between the theoretical cutting force output by the digital twin model and the actual normal cutting force gradually decreases. The calibration process is iterated repeatedly until the error is less than or equal to the preset error. After calibration, the digital twin model is no longer a purely theoretical model, but an empirical model that integrates physical mechanisms and actual machining data, and can more accurately predict the normal cutting force behavior under the machine tool, tool holder, and tool combination.

[0037] Therefore, this embodiment of the invention uses historical measured data for calibration, adjusting the model parameters to better match the values ​​of the current machining system, thus making the theoretical normal range output by the model closer to the actual machining process. For different machine tools, tool holders, or tool combinations, only a new set of measured data under normal operating conditions needs to be collected to recalibrate the model parameters, without modifying the mechanism formula itself, facilitating its application on different equipment. The theoretical normal range predicted by the calibrated digital twin model has a high confidence level. When using this model to screen normal cutting force characteristic data, it can more accurately remove data that may contain early anomalies, ensuring the purity of the training samples.

[0038] In some embodiments, obtaining normal cutting force feature data using a digital twin model includes: mapping candidate cutting force features to the virtual state space corresponding to the digital twin model; when a candidate cutting force feature falls within the theoretical normal range, marking the corresponding candidate cutting force feature as a normal sample and using it to train a normal cutting state model; wherein, the candidate cutting force features come from measured cutting force data under historical normal cutting conditions or generated data from the digital twin model.

[0039] Specifically, candidate cutting force features refer to a set of cutting force feature data to be determined as to whether they belong to a normal state. These data may originate from measured cutting force data under historical normal cutting conditions, or from theoretical feature data directly generated by the digital twin model. Before being verified to fall within the theoretical normal range, they are temporarily referred to as candidates. The virtual state space refers to a multi-dimensional abstract space defined by the digital twin model, with cutting condition parameters and cutting force feature parameters as coordinate axes. In this space, each point represents a set of cutting conditions and the corresponding theoretical cutting force features. The mapping operation places candidate cutting force features at specific locations in this space for comparison with the theoretical normal range. The theoretical normal range can be the upper and lower limits of a single feature parameter, or it can be a region in the multi-dimensional feature space.

[0040] First, candidate cutting force features are acquired. These features can be obtained from two sources: one is extracted from measured cutting force data under historical normal cutting conditions, and the other is generated directly from theoretical feature data by the digital twin model based on given cutting condition parameters. Regardless of the source, these feature data are collectively referred to as candidate cutting force features before verification. Next, these candidate cutting force features are mapped to the virtual state space corresponding to the digital twin model. Specifically, based on the cutting condition parameters at the time of acquiring the candidate feature, such as depth of cut, feed rate, and spindle speed, the corresponding coordinate position is determined in the virtual state space. Then, the feature values ​​of the candidate feature, such as average cutting force, peak cutting force, and fluctuation amplitude, are used as another dimension of coordinates in this space. After mapping, each candidate cutting force feature becomes a point in the virtual state space.

[0041] Next, it is determined whether the point falls within the theoretically normal range predicted by the digital twin model. The theoretically normal range is calculated by the digital twin model based on the same cutting condition parameters. For example, the model predicts that the average cutting force under this condition should be between 450N and 550N, with a fluctuation range between 20N and 40N. If a candidate feature point simultaneously meets the range conditions of all feature dimensions, it is determined to fall within the theoretically normal range. When a candidate cutting force feature falls within the theoretically normal range, it is marked as a normal sample. These marked normal samples are then used to train the normal cutting state model. In this way, only data verified as truly normal by the digital twin model is included in the training set, excluding data that may contain early anomalies or measurement noise.

[0042] Therefore, this embodiment of the invention, through pre-screening using a digital twin model, ensures that only data within the theoretically normal range is used for training, avoiding the misinterpretation of early abnormal data as normal data for model training. This improves the accuracy and robustness of the normal cutting state model. When historical normal data is insufficient, the digital twin model can directly generate candidate feature data within the theoretically normal range for training, solving the problem of insufficient training samples under new tools, new materials, or new working conditions. Traditional supervised learning methods require manual labeling of normal and abnormal features, while this embodiment of the invention automatically determines whether candidate features are normal through the digital twin model, achieving automatic labeling of training samples and reducing manual costs.

[0043] In some embodiments, dynamically determining the warning threshold includes: inputting the current cutting condition parameters into a digital twin model to obtain the theoretical normal range of normal cutting force characteristics; wherein the theoretical normal range varies with the change of a second preset parameter, the second preset parameter including at least one of depth of cut, feed rate or spindle speed; and determining the warning threshold based on the theoretical normal range.

[0044] Specifically, the second preset parameter refers to a set of cutting condition parameters that affect the theoretical normal range, including at least one of depth of cut, feed rate, or spindle speed. These parameters are input variables of the digital twin model, and the theoretical normal range changes as they change. Depth of cut refers to the depth to which the tool cuts into the workpiece perpendicular to the feed direction. Feed rate refers to the speed at which the tool moves relative to the workpiece in the cutting direction. Spindle speed refers to the number of revolutions per minute of the machine tool spindle.

[0045] First, the actual cutting parameters under the current cutting conditions are obtained, specifically including at least one of the depth of cut, feed rate, or spindle speed. These parameters can be read in real time from the CNC system or preset by the operator. Then, these current cutting parameters are input into the pre-constructed and calibrated digital twin model. Based on its internal cutting force mechanism formula and calibrated coefficients, the digital twin model calculates the theoretical normal range of cutting force characteristics that should be exhibited under normal cutting conditions for this set of cutting parameters. For example, when the depth of cut is 2mm, the feed rate is 0.1mm / r, and the spindle speed is 1000rpm, the digital twin model outputs that the theoretical normal range of the average cutting force is 500N±50N, the peak cutting force does not exceed 650N, and the fluctuation amplitude is between 25N and 45N.

[0046] It is important to note that the theoretical normal range varies with the second preset parameter. When the depth of cut increases from 2mm to 3mm, the average cutting force within the theoretical normal range may change from 500N±50N to 750N±75N; when the feed rate decreases, the fluctuation range will also decrease accordingly. This dynamic characteristic allows the theoretical normal range to adapt to different machining conditions.

[0047] Finally, the warning threshold is determined based on the theoretical normal range obtained above. This can be achieved by multiplying the boundary value of the theoretical normal range by a safety factor, or by using statistical methods. Regardless of the specific method used, the warning threshold is directly related to the theoretical normal range and therefore adaptively adjusts with changes in the second preset parameter.

[0048] Therefore, in this embodiment of the invention, since the warning threshold is determined based on the theoretical normal range that varies with parameters such as cutting depth, feed rate, and spindle speed, the warning threshold automatically changes when machining conditions change, eliminating the need for manual resetting and avoiding frequent false alarms or missed alarms caused by fixed threshold schemes when working conditions change. The theoretical normal range is predicted by a digital twin model based on physical mechanisms, possessing interpretability and stability. Compared to purely data-driven threshold setting methods, this approach can still provide a reasonable theoretical normal range when cutting parameters change significantly or exceed historical data ranges, thus ensuring that the warning threshold remains reasonable.

[0049] In some embodiments, extracting real-time cutting force feature data from real-time cutting force data includes: performing filtering and noise reduction preprocessing on the real-time cutting force data, and extracting at least one real-time cutting force feature data from the preprocessed real-time cutting force data, including average cutting force, peak cutting force, cutting force fluctuation amplitude, cutting force change rate, and multi-directional cutting force synthesis features.

[0050] Specifically, filtering refers to removing high-frequency noise or interference from specific frequency components from real-time cutting force data. Filters can include low-pass filters, band-pass filters, median filters, etc. Denoising preprocessing refers to a series of processing operations performed on the original cutting force signal to eliminate non-cutting force components caused by sensor noise, electromagnetic interference, mechanical vibration, etc., so that the extracted features more accurately reflect the true cutting force behavior. Average cutting force refers to the arithmetic mean of the cutting force values ​​over a sampling period. Peak cutting force refers to the maximum value of the cutting force over a sampling period. Cutting force fluctuation amplitude refers to half or the standard deviation of the difference between the maximum and minimum cutting force values ​​over a sampling period, used to characterize the dynamic change degree of the cutting force. Cutting force change rate refers to the derivative of the cutting force with respect to time, used to characterize the drastic degree of cutting force change. Multi-directional cutting force synthesis features refer to the comprehensive features obtained by vector synthesis of cutting force components in multiple directions, such as the X, Y, and Z directions, such as the magnitude or direction of the resultant force.

[0051] First, real-time cutting force data is acquired. This data consists of multi-dimensional force signals continuously collected by a cutting force measuring tool holder at a fixed sampling frequency. Then, the raw data undergoes filtering and noise reduction preprocessing. Specific operations include: using a low-pass filter to remove high-frequency noise above the cutting frequency; using a notch filter to remove power frequency interference; and using a median filter to eliminate pulse spike noise. The preprocessed data is smoother and more stable, eliminating abnormal fluctuations caused by non-cutting force factors.

[0052] Next, at least one real-time cutting force feature data is extracted from the preprocessed real-time cutting force data. For the average cutting force, the average value of the cutting force is calculated within a preset time window. For machining where the cutting force changes periodically per revolution, data from an integer number of cycles is averaged. For the peak cutting force, the maximum value of the cutting force is found within the same time window. For the cutting force fluctuation amplitude, the standard deviation of the cutting force within the time window is calculated, or half the difference between the maximum and minimum values ​​is calculated. For the cutting force change rate, the first derivative of the cutting force signal is calculated; either the instantaneous change rate at each sampling point or the average change rate within the time window can be calculated. For the multi-directional cutting force synthesis feature, the force components in the X, Y, and Z directions are vector-synthesized to obtain the amplitude of the resultant force. The extracted features can be a single feature or a combination of multiple features, which together constitute the real-time cutting force feature vector for subsequent difference calculation.

[0053] Therefore, this embodiment of the invention removes sensor noise and environmental interference through filtering and denoising preprocessing, making the extracted features more realistically reflect the changes in cutting force itself and avoiding misjudgments caused by noise interference. By extracting multiple features, the coverage of various anomaly types is improved. The original real-time cutting force data is a high-frequency time series with a huge amount of data. Through feature extraction, the time series is transformed into a few statistical features, which significantly reduces the computational burden of subsequent machine learning models and makes real-time early warning possible.

[0054] In some embodiments, after determining the abnormal cutting state based on the degree of difference and the warning threshold and performing warning processing, the method further includes: feeding back the real-time cutting force characteristic data and cutting condition parameters corresponding to the warning processing to the digital twin model and the normal cutting state model, so as to update the preset parameters of the digital twin model and the decision boundary of the normal cutting state model.

[0055] Specifically, preset parameters refer to the coefficients or correction factors that need to be determined through calibration in the digital twin model, such as the specific cutting force coefficient, friction coefficient, and stiffness coefficient. These parameters determine the numerical value of the theoretical cutting force output by the digital twin model. The decision boundary refers to the boundary by which the normal cutting state model divides the normal and abnormal regions in the feature space. Data within the normal region is considered normal by the normal cutting state model, i.e., with low variability; data within the abnormal region is considered abnormal, i.e., with high variability. The decision boundary can be characterized by parameters of the normal cutting state model, such as the reconstruction error threshold of an autoencoder or the support vectors of a single-class support vector machine.

[0056] Once the system completes an early warning process, it indicates that an abnormal state has been detected during the current machining process. At this time, the system automatically records the data related to the early warning, specifically including the real-time cutting force characteristic data corresponding to the triggering of the warning (i.e., the set of feature vectors that caused the difference to exceed the threshold), and the cutting condition parameters at that time, such as depth of cut, feed rate, and spindle speed. Then, this data is fed back to the digital twin model and the normal cutting state model, respectively.

[0057] For digital twin models, feedback data is used to update their preset parameters. Specifically, the real-time cutting force characteristic data at the time of anomaly reflects deviations that occur in actual machining. This is due to inaccuracies in certain coefficients within the digital twin model, such as a smaller specific cutting force coefficient, leading to a lower theoretical normal range and causing the actual normal cutting force to be misjudged as abnormal. By using the abnormal data as a new calibration point, parameter re-identification or online correction algorithms, such as recursive least squares, are employed to adjust the preset parameters of the digital twin model, making the theoretical normal range predicted by the digital twin model closer to reality. Even if the anomaly is indeed a real anomaly, the digital twin model can improve its prediction accuracy by learning from the normal state data before the anomaly occurred.

[0058] For normal cutting state models, feedback data is used to update their decision boundaries. During initial training, these models use only normal samples, and their decision boundaries cover only the normal region. When encountering novel anomalies, the normal cutting state model may fail to correctly identify unseen anomalous patterns. By feeding back confirmed anomalous data points to the model, incremental learning methods can be employed, such as adding memory units to the autoencoder or updating the support vector set of a single-class support vector machine, to adjust the decision boundaries of the normal cutting state model. This allows the model to identify similar anomalies earlier and more accurately in the future.

[0059] Therefore, this embodiment of the invention, by feeding back the data generated during early warning processing to both models, can learn from each abnormal event and continuously improve the prediction accuracy of the digital twin model and the discrimination capability of the normal cutting state model. As various new anomalies are encountered during processing, the models gradually learn the boundaries of abnormal patterns through feedback updates. If the early warning is a false alarm caused by a deviation in the parameters of the digital twin model, the feedback update can correct the parameters of the digital twin model and avoid false alarms in subsequent similar situations. Conversely, if it is a missed alarm, the feedback of abnormal data can also enable the digital twin model to identify similar anomalies in the future.

[0060] In some embodiments, the normal cutting state model is an autoencoder, and the difference is the reconstruction error; or, the normal cutting state model is a single-class support vector machine, and the difference is the distance metric in the feature space.

[0061] Specifically, an autoencoder refers to an unsupervised learning neural network model consisting of an encoder and a decoder. The encoder compresses the input data into a low-dimensional latent space, and the decoder then reconstructs the original input from the latent space representation. The training objective is to reconstruct the input as accurately as possible from the output. After training, the reconstruction error is small for normal data; however, for anomalous data, the reconstruction error increases significantly because the model has never learned anomalous patterns. A single-class support vector machine (SVM) refers to a variant of SVM used for anomaly detection. During training, only normal samples are used to find a hyperplane in the feature space that makes normal samples fall into a compact region as much as possible. During testing, samples falling within this region are judged as normal, and those falling outside are judged as anomalous.

[0062] For autoencoders, reconstruction error refers to the degree of difference between the model output and the input data, measured using mean squared error or mean absolute error. A larger reconstruction error indicates that the input data deviates more from the normal pattern learned during model training. For single-class support vector machines, distance metric refers to the distance from the test sample to the decision hyperplane or to the boundary of the normal sample region. A larger distance indicates that the test sample deviates more from the normal region.

[0063] In some embodiments, an abnormal cutting state is determined based on the degree of difference and an early warning threshold, and an early warning process is performed, including: when the degree of difference is greater than or equal to the early warning threshold, outputting the ratio of the degree of difference to the early warning threshold as the early warning confidence level; and / or, outputting the main feature type that causes the abnormality based on the degree of contribution of each feature component in the real-time cutting force feature data to the degree of difference.

[0064] Specifically, the warning confidence level refers to a numerical indicator used to characterize the reliability of a warning, calculated as the ratio of the difference to the warning threshold. When the difference is exactly equal to the warning threshold, the confidence level is 1; when the difference is significantly greater than the warning threshold, the confidence level is greater than 1, and a larger value indicates higher credibility of the warning; when the difference is slightly higher than the threshold, the confidence level is slightly greater than 1, indicating lower certainty of the warning, possibly a critical situation caused by noise. Feature components refer to each independent dimension in the real-time cutting force feature vector. For example, if the real-time cutting force feature data includes three dimensions: average cutting force, peak cutting force, and fluctuation amplitude, then each of these three dimensions is called a feature component. Contribution level refers to the numerical contribution of each feature component to the difference level. A feature component with a larger contribution level indicates a more significant deviation from the normal pattern, and is the main reason for the increased difference level. The main feature type refers to the feature name corresponding to one or more feature components with the largest contribution level, such as average cutting force or peak cutting force, selected according to the contribution level of each feature component, and output as the main feature type causing the anomaly to the operator.

[0065] When the difference is greater than or equal to the warning threshold, a warning process is triggered. During this process, the warning confidence level can be output. Specifically, the ratio of the difference to the warning threshold is calculated, and this ratio is the warning confidence level, reflecting the severity of the abnormal deviation: the larger the ratio, the further the current state deviates from the normal pattern, the higher the confidence level of the warning, and the higher the priority of emergency handling. This warning confidence level can be output together with the warning signal.

[0066] It can also output the main feature types that cause anomalies. Specifically, it analyzes each feature component in the real-time cutting force feature data and calculates the contribution of each feature component to the degree of difference. The method for calculating the degree of contribution can be determined according to the machine learning model used. For example, for an autoencoder, the contribution of each feature dimension can be obtained by calculating the squared difference between the input and output; for a single-class support vector machine, the contribution can be obtained by calculating the partial derivative of the decision function with respect to each feature component. After obtaining the degree of contribution of each feature component, it sorts them from largest to smallest, selects one or more feature components with the largest contribution, and outputs their corresponding feature types, such as average cutting force, as the main feature type.

[0067] Therefore, the warning confidence level in this embodiment of the invention helps operators judge the urgency and reliability of the current warning, avoiding the same shutdown response for every warning. For warnings with low confidence, operators can choose to observe for a period of time; for warnings with high confidence, they should be dealt with immediately. By outputting the main characteristic types causing the anomaly, operators can quickly understand the possible causes of the anomaly, helping them to take targeted measures and reduce troubleshooting time.

[0068] In some embodiments, after performing early warning processing based on the degree of difference and the early warning threshold, the method further includes: when the duration of the degree of difference being greater than or equal to the early warning threshold is greater than or equal to a preset duration, an adjustment command is automatically generated and sent to the CNC system based on the theoretical normal range of the current cutting condition output by the digital twin model. The adjustment command is used to reduce the feed rate or reduce the depth of cut so that the real-time cutting force approaches the theoretical normal range.

[0069] Specifically, the preset duration refers to a pre-defined time threshold used to determine whether a state where the difference exceeds the warning threshold has persisted for a sufficiently long time. The specific value of the preset duration can be set according to the dynamic characteristics of the machining process, for example, it can be set to several sampling cycles or several tool rotation cycles. This parameter is used to filter out brief, sporadic noise interference and avoid unnecessary automatic adjustments triggered by instantaneous spikes. The adjustment command refers to the control command automatically generated by the system and sent to the CNC system to change the machining parameters of the machine tool. The specific content of the adjustment command can be an instruction to reduce the feed rate or a command to reduce the depth of cut. The CNC system refers to the control core of the CNC machine tool, responsible for parsing and executing the machining program, controlling the servo drive system, monitoring the machining status, etc. Approaching refers to the process in which the real-time cutting force, under the action of the adjustment command, gradually approaches the theoretical normal range and eventually falls into that range. Due to the inertia of actual control, it is not required that the real-time cutting force immediately equal the theoretical value, as long as the direction converges towards the normal range and eventually stabilizes within the range.

[0070] Therefore, traditional early warning systems only issue alarm signals, leaving the decision of whether to stop the machine or adjust parameters to the operator, resulting in a delayed response and reliance on manual judgment. This invention, however, automatically generates adjustment instructions after an alarm and sends them to the CNC system, achieving an automated closed loop from anomaly detection to proactive intervention, significantly shortening response time. When the anomaly is not severe enough to require a shutdown, automatically adjusting the feed rate or depth of cut can quickly bring machining back to normal, avoiding production interruptions and workpiece scrap due to downtime. Through continuous monitoring and automatic adjustment, the cutting load can be proactively reduced before serious faults such as tool overload and chipping occur, effectively reducing the risk of major accidents such as tool breakage and spindle damage.

[0071] Figure 2 This is a schematic flowchart illustrating a force measurement and early warning method for a cutting force measuring tool holder provided in an embodiment of the present invention. Based on the existing real-time acquisition of cutting force data and the construction of a digital twin model of the cutting process by the cutting force measuring tool holder, this method introduces machine learning algorithms to model the cutting force behavior, achieving intelligent identification and early warning of abnormal cutting states. The core of this method lies in using the digital twin model as a reference system for normal cutting states and the machine learning algorithm as a tool for identifying state deviations, achieving intelligent early warning of abnormal cutting states through the collaboration of both. Figure 2 As shown, the force measurement early warning method for cutting force measuring tool holders specifically includes the following steps: S201. Cutting force data acquisition and digital twin model construction.

[0072] Specifically, the first step is to acquire raw multidimensional cutting force data using a cutting force-measuring tool holder. This tool holder integrates a strain gauge or piezoelectric multidimensional force sensor. During machining, the tool holder senses the cutting force acting on the tool in real time and outputs analog or digital force signals in three orthogonal directions at a fixed sampling frequency. These signals are then read by a data acquisition card to obtain the raw multidimensional cutting force data. The cutting force data can include unidirectional or multidirectional cutting force components, such as simultaneously acquiring force components in the X, Y, and Z directions.

[0073] Secondly, a digital twin model of normal cutting conditions is constructed based on tool structure parameters, cutting condition parameters, and measured cutting force data under historical normal cutting conditions. First, tool structure parameters and cutting condition parameters are obtained. Then, an initial digital twin model is established using the cutting force mechanism formula. Next, using multiple sets of measured cutting force data collected from past machining of similar workpieces under historical normal cutting conditions, the least squares method is used to calibrate model parameters such as the specific cutting force coefficient and friction coefficient in the initial model, ensuring that the error between the theoretical cutting force output by the digital twin model and the actual normal cutting force is less than or equal to a preset error. After calibration, this digital twin model can describe the dynamic behavior characteristics of cutting force changes with time and conditions under normal cutting conditions, given any set of cutting condition parameters.

[0074] S202, Cutting force feature extraction and state mapping.

[0075] Specifically, the real-time cutting force data is first preprocessed. During machining, real-time cutting force data output from the cutting force measuring tool holder is continuously received. The raw signal is first preprocessed, resulting in smoother and more stable data that accurately reflects the dynamic changes in cutting force. Secondly, core feature parameters such as average cutting force, peak cutting force, cutting force fluctuation amplitude, cutting force change rate, and multi-directional cutting force synthesis characteristics are extracted. Finally, these feature parameters are mapped to a digital twin virtual state space. The extracted real-time cutting force feature vector is combined with the cutting condition parameters at the time of data acquisition and mapped to the virtual state space defined by the digital twin model. This virtual state space is a multi-dimensional space with cutting condition parameters and cutting force feature parameters as coordinate axes. After mapping, each set of real-time data corresponds to a point in the virtual state space. The position of this point contains both current operating condition information and measured feature information, facilitating subsequent comparison with the theoretical normal range predicted by the digital twin model.

[0076] S203. Modeling of normal cutting state based on machine learning.

[0077] Specifically, the normal cutting force feature data generated by the digital twin model is first selected as training samples. Before formal machining begins, model training is performed. Under the condition that the machine tool is in normal cutting conditions, cutting force data is collected for a period of time, and features are extracted. Then, the constructed digital twin model is used to filter each candidate feature data, that is, the candidate features are mapped to a virtual state space to determine whether they fall within the theoretically normal range predicted by the digital twin model. Only candidate features falling within the theoretically normal range are marked as normal samples for subsequent training. Furthermore, when historical normal data is insufficient, the digital twin model can also directly generate theoretical feature data that conforms to the theoretically normal range as training samples.

[0078] Secondly, a normal cutting force behavior representation model is constructed using machine learning algorithms. All selected normal samples are used as training data, and mean squared error is employed as the loss function to train an autoencoder capable of reconstructing the input normal feature vector with high accuracy. When normal features are input, the reconstruction error is small; however, when features deviate from the normal pattern are input, the reconstruction error increases significantly.

[0079] S204. Identification and early warning of abnormal cutting conditions.

[0080] Specifically, firstly, the real-time processed cutting force features are input into the machine learning model. During actual machining, real-time cutting force feature vectors are extracted and input into the trained normal cutting state model. Secondly, the difference between the model output and the input features is calculated to determine the degree of deviation from the normal cutting state. The autoencoder reconstructs the input features based on the learned normal pattern and outputs a reconstructed vector. The mean square error between the input and output vectors is calculated as the difference. This difference quantifies the degree of deviation between the current cutting state and the normal cutting state: the smaller the difference, the closer the current state is to normal; the larger the difference, the more serious the deviation. Finally, when the difference exceeds the warning threshold, it is determined that there is an abnormal risk such as tool overload, abnormal wear, or machining instability.

[0081] S205, Early warning threshold adaptive adjustment.

[0082] Specifically, the warning threshold is dynamically adjusted based on the prediction results of the digital twin model for different cutting conditions. The warning threshold is not a fixed value but rather adaptively adjusts as the cutting condition parameters change. Specifically, the current cutting condition parameters are input into the digital twin model, which outputs the theoretical normal range of the normal cutting force characteristics under that condition. The warning threshold is dynamically calculated based on the width of this theoretical normal range and the statistical distribution of differences among normal samples. This adaptive adjustment mechanism avoids false alarms or missed alarms when cutting parameters change significantly, improving the reliability and stability of the warning judgment.

[0083] S206, Early Warning Output and Model Update.

[0084] Specifically, the system first outputs an anomaly warning signal to indicate machining risks. Upon determining an anomaly risk, warning actions are immediately executed. Furthermore, warning information, including the degree of difference, warning threshold, warning confidence level, and the main characteristic type causing the anomaly, can be sent to the CNC system or cloud platform for decision-making by operators or upper management systems. Secondly, anomaly data is fed back to the digital twin model and machine learning model for continuous optimization. After each warning action, real-time cutting force characteristic data, cutting condition parameters, and degree of difference information at the time the warning was triggered are automatically recorded. During machining breaks or periodic maintenance, this feedback data is used to update both models. Through this continuous optimization, the system adapts to new machining conditions and anomaly types over long-term operation.

[0085] This invention also provides a force measurement early warning device for a cutting force-measuring tool holder. Figure 3 This is a schematic diagram of the force measurement early warning device for a cutting force measuring tool holder provided in an embodiment of the present invention. Figure 3 As shown, the force measurement and early warning device for the cutting force measuring tool holder includes a twin model construction module 301, used to construct a digital twin model of the cutting process based on a first preset parameter; wherein, the digital twin model is used to describe the dynamic behavior of the cutting force under normal cutting conditions as the cutting condition parameters change, and the first preset parameter includes tool structure parameters, cutting condition parameters, and measured cutting force data under historical normal cutting conditions; a machine learning training module 302 is used to train a normal cutting state model under normal cutting conditions using the normal cutting force feature data obtained from the digital twin model as training samples; wherein, the normal cutting state The model is used to reconstruct or fit normal cutting force characteristic data; the difference parameter calculation module 303 is used to input the real-time cutting force characteristic data extracted from the real-time cutting force data into the normal cutting state model, and calculate the difference between the real-time cutting force characteristic information output by the normal cutting state model and the real-time cutting force characteristic data; the dynamic early warning processing module 304 is used to determine the abnormal cutting state based on the difference and the early warning threshold, and perform early warning processing; wherein, the early warning threshold is dynamically determined based on the theoretical normal range predicted by the digital twin model for the normal cutting force behavior under the current cutting condition.

[0086] This invention also provides an electronic device. Figure 4 This is a schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention. Figure 4As shown, the electronic device may include: a processor 401, a communication interface 402, a memory 403, and a communication bus 404, wherein the processor 401, the communication interface 402, and the memory 403 communicate with each other via the communication bus 404. The processor 401 can call logical instructions in the memory 403 to execute a force measurement early warning method for a cutting force-measuring tool holder, including: A digital twin model of the cutting process is constructed based on a first preset parameter; wherein, the digital twin model is used to describe the dynamic behavior of the cutting force under normal cutting conditions as the cutting condition parameters change, and the first preset parameter includes tool structure parameters, the cutting condition parameters, and measured cutting force data under historical normal cutting conditions; Under normal cutting conditions, the normal cutting force feature data obtained by the digital twin model is used as training samples to train a normal cutting state model; wherein, the normal cutting state model is used to reconstruct or fit the normal cutting force feature data. The real-time cutting force feature data extracted from the real-time cutting force data is input into the normal cutting state model, and the difference between the real-time cutting force feature information output by the normal cutting state model and the real-time cutting force feature data is calculated. An abnormal cutting state is determined based on the difference and the warning threshold, and a warning is issued; wherein, the warning threshold is dynamically determined based on the theoretical normal range predicted by the digital twin model for the normal cutting force behavior under the current cutting condition.

[0087] Furthermore, the logical instructions in the aforementioned memory 403 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0088] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute a force measurement early warning method for cutting a force-measuring tool holder, including: A digital twin model of the cutting process is constructed based on a first preset parameter; wherein, the digital twin model is used to describe the dynamic behavior of the cutting force under normal cutting conditions as the cutting condition parameters change, and the first preset parameter includes tool structure parameters, the cutting condition parameters, and measured cutting force data under historical normal cutting conditions; Under normal cutting conditions, the normal cutting force feature data obtained by the digital twin model is used as training samples to train a normal cutting state model; wherein, the normal cutting state model is used to reconstruct or fit the normal cutting force feature data. The real-time cutting force feature data extracted from the real-time cutting force data is input into the normal cutting state model, and the difference between the real-time cutting force feature information output by the normal cutting state model and the real-time cutting force feature data is calculated. An abnormal cutting state is determined based on the difference and the warning threshold, and a warning is issued; wherein, the warning threshold is dynamically determined based on the theoretical normal range predicted by the digital twin model for the normal cutting force behavior under the current cutting condition.

[0089] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described force measurement early warning method for a cutting force-measuring tool holder, comprising: A digital twin model of the cutting process is constructed based on a first preset parameter; wherein, the digital twin model is used to describe the dynamic behavior of the cutting force under normal cutting conditions as the cutting condition parameters change, and the first preset parameter includes tool structure parameters, the cutting condition parameters, and measured cutting force data under historical normal cutting conditions; Under normal cutting conditions, the normal cutting force feature data obtained by the digital twin model is used as training samples to train a normal cutting state model; wherein, the normal cutting state model is used to reconstruct or fit the normal cutting force feature data. The real-time cutting force feature data extracted from the real-time cutting force data is input into the normal cutting state model, and the difference between the real-time cutting force feature information output by the normal cutting state model and the real-time cutting force feature data is calculated. An abnormal cutting state is determined based on the difference and the warning threshold, and a warning is issued; wherein, the warning threshold is dynamically determined based on the theoretical normal range predicted by the digital twin model for the normal cutting force behavior under the current cutting condition.

[0090] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0091] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A force pre-warning method for a cutting force tool holder, characterized in that, include: A digital twin model of the cutting process is constructed based on a first preset parameter; wherein, the digital twin model is used to describe the dynamic behavior of the cutting force under normal cutting conditions as the cutting condition parameters change, and the first preset parameter includes tool structure parameters, the cutting condition parameters, and measured cutting force data under historical normal cutting conditions; Under normal cutting conditions, the normal cutting force feature data obtained by the digital twin model is used as training samples to train a normal cutting state model; wherein, the normal cutting state model is used to reconstruct or fit the normal cutting force feature data. The real-time cutting force feature data extracted from the real-time cutting force data is input into the normal cutting state model, and the difference between the real-time cutting force feature information output by the normal cutting state model and the real-time cutting force feature data is calculated. An abnormal cutting state is determined based on the difference and the warning threshold, and a warning is issued; wherein, the warning threshold is dynamically determined based on the theoretical normal range predicted by the digital twin model for the normal cutting force behavior under the current cutting condition.

2. The force measurement and early warning method for a cutting force measuring tool holder according to claim 1, characterized in that, A digital twin model of the cutting process is constructed based on the first preset parameters, including: Based on the tool structure parameters and the cutting condition parameters, an initial digital twin model is constructed using the cutting force mechanism formula; The model parameters of the initial digital twin model are calibrated using the measured cutting force data under the historical normal cutting conditions, so that the error between the theoretical cutting force output by the digital twin model and the actual normal cutting force is less than or equal to a preset error.

3. The force measurement and early warning method for a cutting force measuring tool holder according to claim 1, characterized in that, The digital twin model is used to obtain normal cutting force characteristic data, including: The candidate cutting force features are mapped to the virtual state space corresponding to the digital twin model; When the candidate cutting force feature falls within the theoretical normal range, the corresponding candidate cutting force feature is marked as a normal sample and used to train the normal cutting state model; wherein, the candidate cutting force feature comes from the measured cutting force data under historical normal cutting conditions or the generated data of the digital twin model.

4. The force measurement and early warning method for a cutting force measuring tool holder according to claim 1, characterized in that, Dynamically determining the warning threshold includes: The current cutting condition parameters are input into the digital twin model to obtain the theoretical normal range of the normal cutting force characteristics; wherein, the theoretical normal range varies with the change of a second preset parameter, the second preset parameter including at least one of depth of cut, feed rate or spindle speed; The warning threshold is determined based on the theoretical normal range.

5. The force measurement and early warning method for a cutting force measuring tool holder according to claim 1, characterized in that, Extract real-time cutting force characteristic data from real-time cutting force data, including: The real-time cutting force data is filtered and denoised, and at least one real-time cutting force feature data is extracted from the preprocessed real-time cutting force data, including average cutting force, peak cutting force, cutting force fluctuation amplitude, cutting force change rate, and multi-directional cutting force synthesis feature.

6. The force measurement and early warning method for a cutting force measuring tool holder according to claim 1, characterized in that, After determining the abnormal cutting state based on the aforementioned difference and warning threshold, and performing warning processing, the process further includes: The real-time cutting force characteristic data and cutting condition parameters corresponding to the early warning processing are fed back to the digital twin model and the normal cutting state model to update the preset parameters of the digital twin model and the decision boundary of the normal cutting state model.

7. The force measurement early warning method for a cutting force measuring tool holder according to any one of claims 1-6, characterized in that, The normal cutting state model is an autoencoder, and the difference is the reconstruction error; or, the normal cutting state model is a single-class support vector machine, and the difference is the distance metric in the feature space.

8. The force measurement early warning method for a cutting force measuring tool holder according to any one of claims 1-6, characterized in that, Based on the aforementioned difference and warning threshold, an abnormal cutting state is determined, and warning processing is performed, including: When the degree of difference is greater than or equal to the warning threshold, the ratio of the degree of difference to the warning threshold is output as the warning confidence level; and / or, the main feature type causing the anomaly is output according to the degree of contribution of each feature component in the real-time cutting force feature data to the degree of difference.

9. A force-measuring early warning device for a cutting force-measuring tool holder, characterized in that, include: The twin model construction module is used to construct a digital twin model of the cutting process based on a first preset parameter; wherein, the digital twin model is used to describe the dynamic behavior of the cutting force under normal cutting conditions as the cutting condition parameters change, and the first preset parameter includes tool structure parameters, the cutting condition parameters, and measured cutting force data under historical normal cutting conditions; The machine learning training module is used to train a normal cutting state model under normal cutting conditions by using the normal cutting force feature data obtained from the digital twin model as training samples; wherein, the normal cutting state model is used to reconstruct or fit the normal cutting force feature data. The difference parameter calculation module is used to input the real-time cutting force feature data extracted from the real-time cutting force data into the normal cutting state model, and calculate the difference between the real-time cutting force feature information output by the normal cutting state model and the real-time cutting force feature data. The dynamic early warning processing module is used to determine the abnormal cutting state based on the difference degree and the early warning threshold, and to perform early warning processing; wherein, the early warning threshold is dynamically determined based on the theoretical normal range of the normal cutting force behavior predicted by the digital twin model under the current cutting conditions.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the force measurement early warning method for the cutting force measuring tool holder as described in any one of claims 1-8 through the computer program.

Images