Numerical control machine tool spindle chip abnormality detection method, device and medium

Abstract

The application discloses a numerical control machine tool spindle chip clamping abnormality detection method, equipment and medium, relates to the numerical control machine tool detection field, and the method comprises the following steps: collecting the tool parameter corresponding to the current tool on the machine tool spindle and the current working condition parameter, and collecting the vibration data corresponding to the current tool idling; the neural network model outputted by pre-training outputs the adaptive threshold value corresponding to the current tool; the chip clamping abnormality of the current tool is detected; wherein, in the training stage, the corresponding reference threshold value is generated and stored by extracting normal idling data; in the reasoning stage, the reference threshold value is corrected by the tool parameter and the working condition parameter. The defects that the traditional fixed threshold method is difficult to adapt to tool differences and working condition changes are solved, the alarm line is dynamically adjusted according to the current specific conditions, and the false alarm and missed alarm risk caused by normal background vibration changes is reduced.

Description

Technical Field

[0001] This application relates to the field of CNC machine tool testing, specifically to methods, equipment, and media for detecting abnormal chip clamping in CNC machine tool spindles. Background Technology

[0002] Computer Numerical Control (CNC) machine tools, often simply called numerical control machine tools, present a challenging and destructive process anomaly during high-efficiency and precision machining. This refers to the phenomenon where metal or non-metal chips generated during cutting fail to be effectively flushed away by the coolant or smoothly discharged through the chip removal system. Instead, under the influence of high-speed rotation and complex cutting forces, these chips become entangled, embedded, or adhered within the tool holder (including the critical contact area that mates with the machine tool spindle taper hole) or the clamping mechanism.

[0003] When chips get stuck in the tool holder, the following consequences may occur:

[0004] 1. Chips and foreign objects on the tool holder disrupt the originally precise tapered fit and rigid connection between the tool holder and the machine tool spindle taper hole. This minute displacement or interference causes an uncontrollable slight shift in the tool's clamping position on the spindle or increased vibration, resulting in a loss of machining accuracy.

[0005] 2. Precision deviations caused by chip inclusions often mean that the workpiece does not meet the requirements and may even lead to its scrapping.

[0006] 3. Chip clamping not only affects accuracy but also disrupts the dynamic balance and stability of the tooling system. Continuous abnormal vibration and uneven stress can easily lead to chipping, breakage, or abnormal wear of the cutting tool, causing damage to both the tool and the equipment. Summary of the Invention

[0007] To address the aforementioned problems, this application proposes a method for detecting chip clamping abnormalities in CNC machine tool spindles, comprising:

[0008] The tool parameters and current working condition parameters corresponding to the current tool on the machine tool spindle are collected, and vibration data corresponding to the current tool idling is collected based on the vibration sensor installed on the CNC machine tool spindle;

[0009] Based on the tool parameters, working condition parameters, and vibration data as inputs, the adaptive threshold corresponding to the current tool is output through a pre-trained neural network model.

[0010] By comparing the vibration data with the adaptive threshold, chip clamping anomalies are detected in the current tool.

[0011] In the training phase, the neural network model generates and stores a corresponding baseline threshold by extracting normal idle data; in the inference phase, it corrects the baseline threshold by using the tool parameters and working condition parameters corresponding to the current tool to generate an adaptive threshold corresponding to the current tool, and uses the adaptive threshold to detect chip clamping anomalies in the vibration data of the current tool.

[0012] In one example, the neural network model includes the following modules:

[0013] The decoupling module, a universal module for both the training and inference phases, performs coordinate transformation on the collected three-dimensional acceleration through the spindle angle of the machine tool spindle and separates the radial acceleration sensitive to chip clamping.

[0014] The resonance enhancement module, a dedicated module for the training phase, extracts the fundamental energy of the radial acceleration through bandpass filtering and determines the harmonic distortion index.

[0015] The feature distribution modeling module is a dedicated module for the training phase. It performs parameter fitting based on normal idling data and Gaussian mixture model to obtain the joint probability distribution model corresponding to the fundamental energy and the harmonic distortion index, and obtains the corresponding mixed probability distribution parameters.

[0016] The benchmark threshold generation module is a dedicated module for the training phase. It generates benchmark vibration features corresponding to each data type in the tool parameters and working condition parameters based on the mixed probability distribution parameters, and fuses the benchmark vibration features to obtain the benchmark threshold.

[0017] The monitoring interface module is a dedicated module for the inference phase. Based on the tool parameters and working condition parameters corresponding to the current tool, it corrects the stored benchmark threshold, generates an adaptive threshold corresponding to the current tool, and uses the adaptive threshold to detect chip clamping anomalies in the radial acceleration corresponding to the vibration data of the current tool.

[0018] In one example, the decoupling module determines that the input data includes the three-dimensional acceleration and principal axis angle in the vibration data; the three-dimensional acceleration includes X-axis acceleration, Y-axis acceleration, and Z-axis acceleration.

[0019] The three-dimensional acceleration in the continuous time domain is discretized according to the angle intervals corresponding to the principal axis angles to obtain the average acceleration value corresponding to each angle interval.

[0020] Divide the interval according to the angle, decompose the mean acceleration into a rotating coordinate system, and obtain the radial acceleration and tangential acceleration corresponding to the machine tool spindle;

[0021] The radial acceleration is retained, while the tangential acceleration is discarded.

[0022] In one example, the resonance enhancement module, for the input sample data, determines that the sample data includes the radial acceleration and the spindle speed of the machine tool spindle included in the operating parameters;

[0023] The fundamental frequency of spindle rotation is determined by the spindle speed, and the in-phase and orthogonal components of the radial acceleration sequence are extracted based on orthogonal projection to obtain the fundamental energy.

[0024] The harmonic distortion index is obtained based on the fundamental wave energy and the corresponding harmonic energy.

[0025] In one example, the feature distribution modeling module determines, for the input sample data, that the sample data includes the fundamental energy, the harmonic distortion index, the tool parameters, and the working condition parameters;

[0026] Based on the fundamental wave energy and the harmonic distortion index, sample data carrying outliers are removed to obtain sample data representing normal idling data.

[0027] Select the corresponding data type for the tool parameters and the working condition parameters, and generate the corresponding working condition label according to the data type. Establish the joint probability distribution model corresponding to the fundamental energy and the harmonic distortion index through the Gaussian mixture model, and obtain the corresponding mixed probability distribution parameters according to the joint probability distribution model.

[0028] In one example, the baseline threshold generation module determines, for the input sample data, that the sample data includes the mixed probability distribution parameters, the tool parameters, and the working condition parameters;

[0029] Select the corresponding data type for the tool parameters and the working condition parameters, generate the corresponding working condition label according to the data type, and extract the radial vibration features corresponding to each working condition label according to the mixed probability distribution parameter.

[0030] For each data type, the reference vibration characteristics corresponding to the data type are determined based on the radial vibration characteristics corresponding to each data type.

[0031] Based on the reference vibration characteristics corresponding to each data type, the reference thresholds corresponding to all data types are obtained by fusion.

[0032] In one example, the monitoring interface module stores the benchmark threshold generated by the benchmark threshold generation module;

[0033] The tool parameters and working condition parameters corresponding to the current tool are compared with their respective preset conventional thresholds to obtain the corresponding parameter ratios;

[0034] Based on the parameter ratio, the stored benchmark threshold is corrected to generate an adaptive threshold corresponding to the current tool; and the radial acceleration of the vibration data corresponding to the current tool is generated through the decoupling module.

[0035] Chip-clamping anomalies are detected by comparing the radial acceleration with the adaptive threshold.

[0036] In one example, the method further includes:

[0037] The results of abnormal chip clamping detection over a period of time were sampled and statistically analyzed, and the model parameters of the neural network model were fine-tuned based on the statistical results.

[0038] On the other hand, this application also proposes a chip clamping abnormality detection device for CNC machine tool spindles, comprising:

[0039] At least one processor; and,

[0040] A memory communicatively connected to the at least one processor; wherein,

[0041] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the CNC machine tool spindle chip clamping anomaly detection method as described in any of the above examples.

[0042] On the other hand, this application also proposes a non-volatile computer storage medium storing computer-executable instructions, which configure the CNC machine tool spindle chip clamping anomaly detection method as described in any of the above examples.

[0043] The method for detecting chip clamping abnormalities in CNC machine tool spindles proposed in this application can bring the following beneficial effects:

[0044] 1. By combining the current tool parameters and working condition parameters with a neural network model, the stored historical benchmark thresholds are corrected in real time to generate adaptive thresholds. This solves the problem that traditional fixed threshold methods are difficult to adapt to differences in tools and changes in working conditions. This application generates different alarm thresholds for different tools and dynamically adjusts the alarm line for the current tool to reduce the risk of false alarms and missed alarms caused by normal mechanical vibration changes.

[0045] 2. Vibration data is collected and detected during the tool's idle period, focusing on the idle phase, which occurs before actual cutting begins. This enables early non-destructive testing and avoids the risk of damage to the workpiece being machined. Furthermore, there is no cutting force interference during idle, and the vibration signal mainly reflects the state of the tool holder-spindle system. Abnormal vibration signals caused by chip clamping are more easily identified, improving the signal-to-noise ratio of the detection.

[0046] 3. During the inference phase, input data is processed in real time through a lightweight interface, ensuring the efficiency of threshold calculation and comparison processes during the inference phase, meeting the machining cycle requirements of CNC machine tools, achieving low-latency response, and reducing interference with normal machining processes.

[0047] 4. The neural network model uses a decoupling module to separate the radial acceleration sensitive to chip clamping based on coordinate transformation, effectively filtering out spindle angle interference and improving the specificity of vibration signal characteristics;

[0048] By using the resonance enhancement module, the fundamental wave energy and harmonic distortion index are extracted to enhance the resonance characteristics caused by the chip inclusions and improve the identification of abnormal signals.

[0049] The feature distribution modeling module establishes a joint probability distribution of normal idling based on a Gaussian mixture model, accurately describing the distribution law of vibration characteristics and providing a statistical basis for threshold generation.

[0050] The reference threshold generation module integrates tool parameters and working condition parameters to generate dynamic reference thresholds, enabling adaptive reference setting under different machining conditions.

[0051] The monitoring interface module can be used to correct the stored baseline thresholds in real time to adapt to the current tooling conditions, ensuring that the detection thresholds match the actual conditions and improving the accuracy of anomaly detection. Attached Figure Description

[0052] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0053] Figure 1 This is a flowchart illustrating the method for detecting chip clamping abnormalities in the spindle of a CNC machine tool in this embodiment of the application.

[0054] Figure 2 This is a schematic diagram of the installation of the vibration sensor in one scenario according to an embodiment of this application;

[0055] Figure 3 In one scenario described in this application, Figure 2 A schematic diagram of the AA cross-section;

[0056] Figure 4In one scenario described in this application, Figure 3 A magnified view of a portion at point C;

[0057] Figure 5 In one scenario described in this application, Figure 2 BB cross-sectional diagram;

[0058] Figure 6 In one scenario described in this application, Figure 5 A magnified view of a portion at point D;

[0059] Figure 7 This is an architecture diagram of a chip-clamping anomaly detection system under one scenario in an embodiment of this application.

[0060] Figure 8 This is a schematic diagram of the modules of a neural network model in one scenario of this application embodiment;

[0061] Figure 9 This is a schematic diagram of the chip clamping anomaly detection device for CNC machine tool spindles in an embodiment of this application;

[0062] Among them, 1. main shaft, 2. vibration sensor. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0064] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0065] like Figure 1 As shown in the embodiment of this application, a method for detecting chip clamping abnormalities in a CNC machine tool spindle is provided, including:

[0066] S101: Collect the tool parameters and current working condition parameters corresponding to the current tool on the machine tool spindle, and collect the vibration data corresponding to the current tool when it is idling based on the vibration sensor installed on the CNC machine tool.

[0067] Different cutting tools have different tool parameters, which may include tool weight, tool length, tool type code, tool holder taper type, chip flute depth, rated speed, etc. Tool parameters can be automatically obtained by the machine tool from a tool information database, or they can be manually entered.

[0068] In actual processing, different tool types and corresponding machining methods may have different chip tolerance values ​​depending on the usage requirements. For example, finishing requires high precision, so the chip tolerance value is relatively small; while in roughing, even if some chips are trapped at the tool holder, the impact on machining accuracy is within an acceptable range. Therefore, to facilitate recording and subsequent threshold modeling, the machine tool is numbered and labeled with different tools, each tool having a unique number, and the corresponding record is made in the database, thereby enabling the management and inspection of the tool.

[0069] Simultaneously, current operating parameters are collected, including various types (referred to here as data types). For example, data types may include temperature status and spindle speed. Temperature status describes whether the current operating condition is cold or hot, as the tool and spindle material may undergo thermal deformation at the current temperature. Spindle speed describes whether the current speed is low or high. Temperature status and spindle speed can be obtained from corresponding sensors on the CNC machine tool.

[0070] In this design, a vibration acceleration sensor is used. The sensor is pre-installed on the casting where the machine tool spindle is located to collect vibration data generated when the tool rotates freely on the spindle. For example... Figures 2-6 As shown, the vibration sensor 2 should be installed as close as possible to the tool holder and on the linkage casting that contacts the tool holder, such as the front or rear face of the spindle 1. Typically, after changing the tool, the tool is allowed to idle, and chip clamping is detected during this idle phase; therefore, vibration data generated during idle operation is collected.

[0071] like Figure 7 As shown, after tool switching, during the idle rotation for chip detection, vibration sensors detect the vibration acceleration values ​​along the X, Y, and Z axes and upload them to the control board via the CAN bus. The X, Y, and Z axes are coordinate axes in the spatial coordinate system established by the CNC machine tool, and their directions are fixed relative to real space. For ease of description, the X-axis acceleration, Y-axis acceleration, and Z-axis acceleration are collectively referred to as three-dimensional acceleration.

[0072] Vibration data can include not only three-dimensional acceleration but also spindle angle. Spindle angle refers to the rotational angle of the machine tool spindle during rotation.

[0073] like Figure 7As shown, the vibration sensor uploads the collected vibration data to the control board, which then processes the data. The control board can receive and process the vibration data from the sensor, and can also store alarm records and adaptive thresholds for different tool parameters on its chip. The adaptive thresholds are used to determine whether the current tool may have chip clamping issues.

[0074] Meanwhile, the control board can also display the working status via indicator lights. When working normally or when a chip-clamping abnormality is detected, the indicator lights will show the corresponding color. The control board can also communicate with the host software via a network cable, and connect to a programmable logic controller (PLC) via an input / output (IO) interface to execute corresponding logic control functions and perform logic control on the CNC machine tool. For example, when a chip-clamping abnormality is detected, the operation of the CNC machine tool can be terminated.

[0075] S102: Based on the tool parameters, the working condition parameters, and the vibration data as inputs, the adaptive threshold corresponding to the current tool is output through a pre-trained neural network model.

[0076] like Figure 7 As shown, the control board collects tool parameters and vibration data, and then transmits them to the host software. The host software stores a pre-trained neural network model, which outputs an adaptive threshold. This adaptive threshold can also be manually adjusted and set within the host software as needed.

[0077] The host software switches between learning and monitoring modes. In learning mode, the neural network model is trained using collected data. In monitoring mode, the collected tool parameters and vibration data are used to output a corresponding adaptive threshold, which is then used to determine if a chip clamping anomaly might exist. If so, the result is sent to the control board, which then executes the appropriate control commands. Alternatively, the host software can directly send the adaptive threshold to the control board, allowing the board to determine the presence of a chip clamping anomaly.

[0078] The host software also allows for visualization, providing real-time display of tool parameters and vibration data, such as real-time display of acceleration vibration curves in the X, Y, and Z axes. It also allows for reviewing historical records of chip clamping detection alarms and resetting control boards in alarm states within the host software.

[0079] S103: By comparing the vibration data with the adaptive threshold, the current tool is subjected to chip clamping anomaly detection.

[0080] Generally, when vibration data exceeds an adaptive threshold, it is considered that the current tool is likely experiencing chip clamping anomalies and requires appropriate processing to remove the chips. For example, when the three-dimensional acceleration in the vibration data exceeds the adaptive threshold, it is considered that there are chip clamping anomalies.

[0081] The adaptive threshold is obtained by the host software through a neural network model, and the comparison between the adaptive threshold and the current vibration data can be implemented in the host software or the control board.

[0082] like Figure 8 As shown, the neural network model includes a decoupling module, a resonance enhancement module, a feature distribution modeling module, a baseline threshold generation module, and a monitoring interface module. The decoupling module is a general module used in both the training and inference phases (i.e., when the model is actually used after training). The resonance enhancement, feature distribution modeling, and baseline threshold generation modules are dedicated to the training phase and are used only during training. The monitoring interface module is dedicated to the inference phase and is used only during inference. Of course, the neural network model may also include an input layer, which will not be discussed further here.

[0083] During the training phase, the neural network model extracts and stores corresponding baseline thresholds from normal idle data. During the inference phase, it corrects these baseline thresholds using the tool parameters and operating condition parameters corresponding to the current tool, generating an adaptive threshold for the current tool. This adaptive threshold is then used to detect chip clamping anomalies in the vibration data of the current tool. The structure and function of each module will be described in detail below.

[0084] In Example 1, the decoupling module is a general-purpose module for both the training and inference phases. Its main function is to synchronously transform the three-dimensional acceleration using the spindle angle and separate the radial vibration component sensitive to chip clamping. While the three-dimensional acceleration can represent the current state of the tool during vibration, the tool itself also causes a certain degree of wobbling during rotation. Therefore, the three-dimensional acceleration detected by the vibration sensor may actually contain a significant amount of noise. By transforming the coordinates using the spindle angle, it is converted into radial and tangential vibration components corresponding to the machine tool spindle, and only the radial vibration component sensitive to chip clamping is selected.

[0085] For the decoupling module, during the training process of the neural network model, its input data consists of the collected sample data. Once training is complete, corresponding to the inference phase, its input data consists of the currently collected actual data. The input data includes the three-dimensional acceleration and principal axis angle from the vibration data; the three-dimensional acceleration includes X-axis acceleration, Y-axis acceleration, and Z-axis acceleration. A corresponding three-dimensional Cartesian coordinate system is generated based on the real space, and the acceleration components along the X, Y, and Z axes are taken as the three-dimensional acceleration. In this case, the three-dimensional acceleration is relative to a fixed direction in the real space.

[0086] The continuous-time-domain three-dimensional acceleration is discretized by dividing the data into intervals based on the principal axis angles, yielding the mean acceleration value for each interval. Since three-dimensional acceleration is a continuously acquired quantity, acquiring it constitutes a continuous-time-domain signal. Here, the X-axis acceleration, Y-axis acceleration, and Z-axis acceleration are referred to as... .

[0087] spindle angle The spindle angle is obtained by the rotational position fed back by the encoder. The spindle angle ranges from 0° to 360°. It is pre-divided into multiple angle intervals, for example, each interval is divided into 10° intervals, resulting in 36 angle intervals.

[0088] Discretization is performed based on the angle intervals corresponding to the principal axis angles, thereby calculating the state of the three-dimensional acceleration in each angle interval, as shown in Formula 1:

[0089] Formula 1;

[0090] in, Let the i-th angle be the starting point for dividing the interval. Define the width corresponding to the pre-set angle interval; for example, set it to 10°. The representation of the interval for the i-th angle; For the p-th sampling time, This represents the real-time principal axis angle corresponding to the p-th sampling time. Let k be the three-dimensional acceleration corresponding to the p-th sampling time. When k = x, y, z, k represents the X-axis acceleration, Y-axis acceleration, and Z-axis acceleration, respectively. The number of samples in the interval divided by the i-th angle is based on the number of collection rounds and the collection frequency in each round. Generally speaking, one rotation of the main axis is considered as one collection round. In each collection round, the faster the collection frequency, the higher the number of samples. The mean acceleration of the three-dimensional acceleration corresponding to the interval divided by the i-th angle.

[0091] This process eliminates the influence of speed fluctuations and transforms non-stationary time-domain signals into stationary angle-domain signals.

[0092] Divide the intervals according to the angle, decompose the mean acceleration into a rotating coordinate system, and obtain the radial acceleration and tangential acceleration corresponding to the machine tool spindle; retain the radial acceleration and discard the tangential acceleration.

[0093] Unlike three-dimensional acceleration and the mean acceleration, radial acceleration and tangential acceleration are relative to the machine tool spindle. Since the direction of the machine tool spindle may change during actual operation, their direction relative to the real space is not fixed, but their direction relative to the machine tool spindle is fixed.

[0094] The process of decomposing a rotating coordinate system involves dividing the coordinate system into intervals by angles, transforming the mean acceleration, and obtaining the radial acceleration and tangential acceleration, as shown in Formula 2:

[0095] Formula 2;

[0096] in, 、 、 These are the mean acceleration values ​​of the three-dimensional acceleration corresponding to the interval divided by the i-th angle; , is the center angle that divides the interval by the i-th angle; The radial acceleration is used to divide the interval at the i-th angle. Divide the interval for tangential acceleration at the i-th angle.

[0097] The radial acceleration obtained at this point represents the component of centrifugal force, directly characterizing the mass imbalance effect caused by centrifugal force due to chip clamping, and is therefore retained as a valid signal. The tangential acceleration, representing the component of torque fluctuation, mainly includes background noise such as bearing friction and is weakly correlated with chip clamping, and is therefore discarded.

[0098] By explicitly separating the centrifugal force effect through rotational coordinate transformation, replacing the traditional black-box feature extraction, the output radial acceleration can be directly used for unbalance calculation and can suppress noise caused by speed fluctuations.

[0099] In Example 2, for the resonance enhancement module, since it is a dedicated module for the training phase, the input is all sample data. For the input sample data, it is determined that the sample data includes radial acceleration. The input data includes the machine tool spindle speed (RPM), and after the angle intervals have been divided, the input data can also include the center angle of the angle intervals. .

[0100] The spindle rotation fundamental frequency is determined by the spindle speed, as shown in Formula 3:

[0101] Formula 3;

[0102] in, The spindle rotational fundamental frequency is denoted by RPM, which is the spindle speed. Since the spindle speed RPM is measured in revolutions per minute (rpm) while the spindle rotational fundamental frequency is measured in revolutions per second (rpm), the spindle rotational fundamental frequency is obtained through proportional conversion.

[0103] By orthogonally projecting the fundamental frequency of the spindle rotation, the in-phase and quadrature components of the radial acceleration sequence are extracted to obtain the fundamental wave energy, which can be expressed as shown in Formula 4:

[0104] Formula 4;

[0105] in, I represents the fundamental frequency energy corresponding to the fundamental frequency of the spindle rotation, Q represents the fundamental frequency in-phase component (referred to as in-phase component), and M represents the fundamental frequency quadrature component (referred to as quadrature component). M represents the number of angle division intervals.

[0106] After expanding the in-phase and quadrature components, we can see the results as shown in Formula 5:

[0107] Formula 5;

[0108] in, The center time of dividing the interval for the i-th angle. Then it is an in-phase component. These are orthogonal components.

[0109] Since the center angle of the angle division interval has already been obtained above... Therefore, Formula 5 can be converted into Formula 6:

[0110] Formula Six;

[0111] in, The central angle for dividing the angle interval can be determined from the correspondence between the angle domain and the time domain. Therefore, Formula 5 can be converted into Formula 6, which makes calculation easier.

[0112] Similar to the fundamental frequency energy, the corresponding harmonic energy is calculated as shown in Formula 7:

[0113] Formula 7;

[0114] Where n=1 is the fundamental frequency energy calculated by this formula, and when n is any other positive integer, the calculated energy is the corresponding nth harmonic energy. For example, when n=2 or 3, the energy is the corresponding second harmonic energy. Third harmonic energy .

[0115] Based on the fundamental frequency energy and the corresponding harmonic energies, the harmonic distortion index is obtained, as shown in Formula 8:

[0116] Formula 8;

[0117] Where HD is the harmonic distortion index. , , These are the fundamental energy, the second harmonic energy, and the third harmonic energy, respectively.

[0118] The obtained fundamental wave energy is specifically sensitive to linear mass imbalance, avoiding noise interference from general frequency domain analysis, and can be used to detect the quality of chip clamping; the harmonic distortion index is specifically sensitive to nonlinear impact, and uses nonlinear dynamic characteristics to distinguish between chip clamping and ordinary imbalance, and can be used to detect the collision intensity of chip clamping, thereby realizing the physical enhancement and noise suppression of chip clamping fault characteristics.

[0119] In Example 3, for the feature distribution modeling module, since it is a dedicated training module, the input sample data includes fundamental energy, harmonic distortion index, tool parameters, and operating parameters. The fundamental energy and harmonic distortion index are output by the resonance enhancement module, while the tool parameters and operating parameters are acquired through machine tool acquisition or manually input. Therefore, each sample data includes the corresponding fundamental energy, harmonic distortion index, tool parameters, and operating parameters.

[0120] For fundamental frequency energy and harmonic distortion index, sample data carrying outliers are removed to obtain sample data representing normal idling data. Since the currently acquired sample data may contain some outliers, which are detrimental to the learning of adaptive thresholds during the current training process, sample data carrying outliers are removed.

[0121] The elimination process is shown in Formula Nine:

[0122] Formula Nine;

[0123] Among them, valid data This refers to the remaining sample data after removing outlier samples. , These are the fundamental wave energies. The median of the harmonic distortion index HD, while The valid data obtained at this point is the sample data representing normal idling data (also known as historical regular idling data).

[0124] Select the corresponding data type for tool parameters and operating condition parameters, and generate corresponding operating condition labels based on the data type. Here, select tool type in the tool parameter data type, and select temperature state and spindle speed as operating condition labels in the operating condition parameter data type. For these three data types, set different preset ranges for generating operating condition labels. For example, label a temperature state below a preset temperature (e.g., 40°C) as cold machine, and label a temperature state above a preset temperature as hot machine; label a spindle speed below a preset spindle speed (e.g., 8000 RPM) as low speed, and label a spindle speed above a preset spindle speed as high speed; label a tool weight below a preset weight (e.g., 0.5 kg) as finishing, and label a tool weight above a preset weight as roughing. At this point, each data type includes two operating condition labels. Of course, in actual work, these can be changed according to requirements, such as adding corresponding parameters or changing the corresponding threshold.

[0125] Based on this, a joint probability distribution model corresponding to the fundamental energy and harmonic distortion index is established through a Gaussian mixture model, and the corresponding mixture probability distribution parameters are obtained from the joint probability distribution model.

[0126] At this point, for the sample data, a corresponding joint probability distribution model is generated, and the corresponding mixture probability distribution parameters are obtained. Let the feature vector set of the sample data be: ,in, Let be the feature vector of the i-th sample data, which is a two-dimensional vector composed of fundamental energy and harmonic distortion index. , Let i be the fundamental energy sequence of the i-th sample data. The harmonic distortion index is set for the i-th sample data, and a corresponding operating condition label is added to each sample data, such as a temperature status label. Spindle speed label Tool type label .

[0127] At this point, the joint probability distribution model adopts a Gaussian mixture model with the probability density function shown in Equation 10:

[0128] Formula 10;

[0129] Wherein, K is the preset number of Gaussian distributions, representing the number of working condition labels, which can be 6 according to the description in the above embodiment; Let be the probability of the k-th working condition label appearing, and its value ranges from [0,1]. ; The center position of the feature vector of the kth working condition label; Let K be the covariance matrix corresponding to the k-th working condition label. The distribution is a single Gaussian distribution, describing the characteristic distribution pattern of the k-th working condition label. The expansion of the single Gaussian distribution is shown in Formula 11:

[0130] Formula 11;

[0131] Where d=2 is the feature dimension. Let be the determinant of the covariance matrix.

[0132] Having constructed the joint probability distribution model, it needs to be solved to obtain the mixed probability distribution parameters. This can be done using the expectation-maximization algorithm, as shown in Equation Twelve:

[0133] Formula 12;

[0134] in, Let be the probability that the i-th sample data belongs to the Gaussian component k, which is the probability that it belongs to the k-th working condition label. It is a soft assignment probability, and the same sample data may belong to multiple working condition labels at the same time.

[0135] Using the obtained probabilities, a maximization step is performed to calculate the parameters of the mixed probability distribution, as shown in Formula 13:

[0136] Formula Thirteen;

[0137] Where N is the total number of sample data. Let be the number of samples in the k-th working condition label. Let be the mixing coefficient of the k-th working condition label, representing the state weight of the k-th working condition label, and characterizing the proportion of each working condition label. Let be the mean vector of the k-th working condition label, representing the center of the feature distribution. Let be the covariance matrix of the k-th operating condition label, representing the joint volatility of the fundamental energy and harmonic distortion index. Through iterative processing using the EM algorithm, the final mixture probability distribution parameters include the mixture coefficients corresponding to each operating condition label. Mean vector Covariance matrix .

[0138] In Example 4, the baseline threshold generation module is a dedicated training module. For the input sample data, it determines that the sample data includes mixed probability distribution parameters, tool parameters, and working condition parameters. Here, the mean vector is selected from the mixed probability distribution parameters. Covariance matrix For display purposes.

[0139] Select the corresponding data type for tool parameters and working condition parameters, and generate the corresponding working condition label based on the data type. The selection and generation methods of the data type and working condition label are the same as those described above, and will not be repeated here.

[0140] Based on the mixed probability distribution parameters, the radial vibration characteristics corresponding to each working condition label are extracted, as shown in Formula Fourteen:

[0141] Formula Fourteen;

[0142] in, This is the reference value for radial vibration corresponding to the k-th operating condition label. For the mean vector Take the first element. Let be the fluctuation intensity of the radial vibration corresponding to the k-th operating condition label. Covariance matrix Take the element in the first row and first column, and calculate its square root. This square root represents the variance of the element in the covariance matrix that represents the variance of the first feature itself.

[0143] In the mean vector In the middle, there is the fundamental energy sequence containing sample data i. Harmonic distortion index When setting this mean vector, the value of the fundamental energy sequence is intentionally placed in the first position of the vector, so the first element is chosen as the benchmark value. Similarly, the covariance matrix represents the variance of the feature itself, and the element in the first row and first column represents the variance of the benchmark value corresponding to the first element of the mean vector, that is, the degree of fluctuation.

[0144] For each data type containing operating condition labels, the reference vibration characteristics corresponding to the data type are determined based on the radial vibration characteristics corresponding to each operating condition label. A data type typically includes multiple operating condition labels; for example, when the data type is temperature state, the operating condition labels include low temperature and high temperature; when the data type is tool type, the operating condition labels include roughing and finishing. In this case, the corresponding reference vibration characteristics for that data type are obtained, as shown in Formula 15:

[0145] Formula 15;

[0146] in, Let be the reference value of radial vibration in the reference vibration characteristic corresponding to the t-th data type. The wave intensity of radial vibration in the reference vibration characteristic corresponding to the t-th data type; The reference values ​​of radial vibration corresponding to the k1th and k2th working condition labels are respectively. The fluctuation intensity of radial vibration corresponding to the k1th and k2th working condition labels respectively, and the k1th and k2th working condition labels belong to the tth data type; The safety baseline factor is adjustable; the default value can be 1.2.

[0147] Formula 15 specifies only two working condition labels for each data type. If more working condition labels are actually set, they can be modified accordingly.

[0148] At this point, by fusing the benchmark value and the wave intensity using Formula Sixteen, the corresponding benchmark vibration characteristics can be obtained, as shown in Formula Sixteen:

[0149] Formula Sixteen;

[0150] in, This represents the reference vibration characteristic corresponding to the t-th data type.

[0151] Based on the reference vibration characteristics corresponding to each data type, the reference thresholds for all data types are fused. For example, as shown in Formula 17, a geometric mean fusion method can be used to achieve a dimensionally unified threshold fusion while preserving the independent physical influence characteristics of each factor, resulting in the final reference threshold.

[0152] Formula 17;

[0153] in, As the baseline threshold, ~ These represent the reference vibration characteristics corresponding to each data type, including temperature state, spindle speed, and tool type. In practical work, Formula 17 can also be adapted to change when the selected data type changes.

[0154] In Example 5, the monitoring interface module, a dedicated module for the inference phase, is required to perform analysis at a relatively fast speed during the actual use of the neural network model. Therefore, it stores the baseline threshold generated by the baseline threshold generation module. Then, it uses the actual situation of the current tool (including tool parameters and working condition parameters) to correct the baseline threshold and obtain an adaptive threshold. In other words, the monitoring interface module corrects the baseline threshold based on the current scenario to obtain an adaptive threshold that is suitable for the current scenario.

[0155] For the current tool and operating condition parameters, compare them with their respective preset conventional thresholds to obtain the corresponding parameter ratios. For example, taking the temperature status in the operating condition parameters as an example, the above text defines temperatures below 40°C and above 40°C as low temperature and high temperature, respectively. Therefore, the corresponding preset conventional threshold can also be 40°C. By proportionally calculating the current temperature in the operating condition parameters to this preset conventional threshold, the corresponding ratio can be obtained. Similarly, the preset conventional threshold corresponding to the spindle speed can be 8000 RPM, and the preset conventional threshold corresponding to the tool weight can be 0.5 kg.

[0156] Based on the parameter ratio, the stored baseline threshold is corrected to generate the adaptive threshold corresponding to the current tool, as shown in Formula 18:

[0157] Formula 18;

[0158] in, , , These are the parameter ratios corresponding to temperature, spindle speed, and tool type, respectively. , , These are the corresponding adjustment parameters, which can be obtained by training the monitoring interface module. For example, after other modules in the neural network model have been trained, a baseline threshold is output. Then, using sample data, the adjustment parameters in the monitoring interface module are trained and learned using this baseline threshold.

[0159] When the temperature and spindle speed increase, the tool and workpiece materials may experience thermal expansion, or the centrifugal force and cutting force of the tool may increase, potentially leading to increased vibration and noise during the cutting process. Although some noise is eliminated during identification by retaining only radial acceleration, radial noise may not be eliminated. Therefore, in this case, positive compensation should be applied to the reference threshold to increase it and reduce the probability of false alarms. , It should be a value greater than or equal to 1.

[0160] The decoupling module generates the radial acceleration of the vibration data corresponding to the current tool. The decoupling module is a general module for both the training and inference phases, so the radial acceleration can also be obtained through it during the inference phase.

[0161] Chip-clamping anomaly detection is performed by comparing radial acceleration with an adaptive threshold. When the radial acceleration reaches the adaptive threshold, it is considered that there may be a chip-clamping anomaly.

[0162] After obtaining the neural network model, the detection results of chip clamping anomalies over a period of time can be sampled and statistically analyzed. For example, the detection results of chip clamping anomalies in the most recent week can be extracted, and the model parameters of the neural network model can be fine-tuned based on the statistical results. For example, the model can be adjusted to learning mode through the host software, and the baseline threshold, adjustment parameters, etc. can be relearned and fine-tuned.

[0163] 1. By combining the current tool parameters and working condition parameters with a neural network model, the stored historical benchmark thresholds are corrected in real time to generate adaptive thresholds. This solves the problem that traditional fixed threshold methods are difficult to adapt to differences in tools and changes in working conditions. The alarm line is dynamically adjusted according to the current specific conditions, reducing the risk of false alarms and missed alarms caused by normal background vibration changes.

[0164] 2. Vibration data is collected and detected during the tool's idle period, focusing on the idle phase, which occurs before actual cutting begins. This enables early non-destructive testing and avoids the risk of damage to the workpiece being machined. Furthermore, there is no cutting force interference during idle, and the vibration signal mainly reflects the state of the tool holder-spindle system. Abnormal vibration signals caused by chip clamping are more easily identified, improving the signal-to-noise ratio of the detection.

[0165] 3. During the inference phase, input data is processed in real time through a lightweight interface, ensuring the efficiency of threshold calculation and comparison processes during the inference phase, meeting the machining cycle requirements of CNC machine tools, achieving low-latency response, and reducing interference with normal machining processes.

[0166] like Figure 9 As shown in the figure, this application embodiment provides a CNC machine tool spindle chip clamping abnormality detection device, including:

[0167] At least one processor; and,

[0168] A memory communicatively connected to the at least one processor; wherein,

[0169] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the CNC machine tool spindle chip clamping anomaly detection method as described in any of the above embodiments.

[0170] This application provides a non-volatile computer storage medium storing computer-executable instructions, which configure the CNC machine tool spindle chip clamping anomaly detection method as described in any of the above embodiments.

[0171] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0172] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0173] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for detecting chip clamping abnormalities in a CNC machine tool spindle, characterized in that, include: The tool parameters and current working condition parameters corresponding to the current tool on the machine tool spindle are collected, and vibration data corresponding to the current tool when it is idling is collected based on the vibration sensor installed on the CNC machine tool. Based on the tool parameters, working condition parameters, and vibration data as inputs, the adaptive threshold corresponding to the current tool is output through a pre-trained neural network model. By comparing the vibration data with the adaptive threshold, chip clamping anomalies are detected in the current tool. During the training phase, the neural network model extracts and stores a corresponding baseline threshold from normal idle data. During the inference phase, it corrects the baseline threshold using the tool parameters and working condition parameters corresponding to the current tool to generate an adaptive threshold for the current tool. The adaptive threshold is then used to detect chip clamping anomalies in the vibration data of the current tool. The neural network model includes the following modules: The decoupling module, a universal module for both the training and inference phases, performs coordinate transformation on the collected three-dimensional acceleration through the spindle angle of the machine tool spindle and separates the radial acceleration sensitive to chip clamping. The resonance enhancement module, a dedicated module for the training phase, extracts the fundamental energy of the radial acceleration through bandpass filtering and determines the harmonic distortion index. The feature distribution modeling module is a dedicated module for the training phase. It performs parameter fitting based on normal idling data and Gaussian mixture model to obtain the joint probability distribution model corresponding to the fundamental energy and the harmonic distortion index, and obtains the corresponding mixed probability distribution parameters. The benchmark threshold generation module is a dedicated module for the training phase. It generates benchmark vibration features corresponding to each data type in the tool parameters and working condition parameters based on the mixed probability distribution parameters, and fuses the benchmark vibration features to obtain the benchmark threshold. The monitoring interface module is a dedicated module for the inference stage. Based on the tool parameters and working condition parameters corresponding to the current tool, it corrects the stored benchmark threshold, generates an adaptive threshold corresponding to the current tool, and uses the adaptive threshold to detect chip clamping anomalies in the radial acceleration corresponding to the vibration data of the current tool. The baseline threshold generation module determines, for the input sample data, that the sample data includes the mixed probability distribution parameters, the tool parameters, and the working condition parameters; Select the corresponding data type for the tool parameters and the working condition parameters, generate the corresponding working condition label according to the data type, and extract the radial vibration features corresponding to each working condition label according to the mixed probability distribution parameter. For each data type, the reference vibration characteristics corresponding to the data type are determined based on the radial vibration characteristics corresponding to each data type. Based on the reference vibration characteristics corresponding to each data type, the reference thresholds corresponding to all data types are obtained by fusing them together. The monitoring interface module stores the benchmark threshold generated by the benchmark threshold generation module. The tool parameters and working condition parameters corresponding to the current tool are compared with their respective preset conventional thresholds to obtain the corresponding parameter ratios; Based on the parameter ratio, the stored benchmark threshold is corrected to generate an adaptive threshold corresponding to the current tool; and the radial acceleration of the vibration data corresponding to the current tool is generated through the decoupling module. Chip-clamping anomalies are detected by comparing the radial acceleration with the adaptive threshold.

2. The method for detecting chip clamping abnormalities in CNC machine tool spindles according to claim 1, characterized in that, The decoupling module determines that the input data includes the three-dimensional acceleration and principal axis angle in the vibration data; the three-dimensional acceleration includes X-axis acceleration, Y-axis acceleration, and Z-axis acceleration. The three-dimensional acceleration in the continuous time domain is discretized according to the angle intervals corresponding to the principal axis angles to obtain the average acceleration value corresponding to each angle interval. Divide the interval according to the angle, decompose the mean acceleration into a rotating coordinate system, and obtain the radial acceleration and tangential acceleration corresponding to the machine tool spindle; The radial acceleration is retained, while the tangential acceleration is discarded.

3. The method for detecting chip clamping abnormalities in CNC machine tool spindles according to claim 1, characterized in that, The resonance enhancement module, based on the input sample data, determines that the sample data includes the radial acceleration and the spindle speed of the machine tool spindle included in the operating parameters; The fundamental frequency of spindle rotation is determined by the spindle speed, and the in-phase and orthogonal components of the radial acceleration sequence are extracted based on orthogonal projection to obtain the fundamental energy. The harmonic distortion index is obtained based on the fundamental wave energy and the corresponding harmonic energy.

4. The method for detecting chip clamping abnormalities in CNC machine tool spindles according to claim 1, characterized in that, The feature distribution modeling module determines, for the input sample data, that the sample data includes the fundamental energy, the harmonic distortion index, the tool parameters, and the working condition parameters; Based on the fundamental wave energy and the harmonic distortion index, sample data carrying outliers are removed to obtain sample data representing normal idling data. Select the corresponding data type for the tool parameters and the working condition parameters, and generate the corresponding working condition label according to the data type. Establish the joint probability distribution model corresponding to the fundamental energy and the harmonic distortion index through the Gaussian mixture model, and obtain the corresponding mixed probability distribution parameters according to the joint probability distribution model.

5. The method for detecting chip clamping abnormalities in CNC machine tool spindles according to claim 1, characterized in that, The method further includes: The results of abnormal chip clamping detection over a period of time were sampled and statistically analyzed, and the model parameters of the neural network model were fine-tuned based on the statistical results.

6. A CNC machine tool spindle chip clamping anomaly detection device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the CNC machine tool spindle chip clamping anomaly detection method as described in any one of claims 1 to 5.

7. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions set up the CNC machine tool spindle chip clamping anomaly detection method as described in any one of claims 1 to 5.

Images