A method for monitoring the condition of CNC machine tools based on multi-window feature registers

Abstract

This invention provides a method for monitoring the condition of CNC machine tool tools based on multi-window feature registration, relating to the fields of machining and intelligent manufacturing. The invention first acquires tool vibration signals, calibrates time windows at fixed intervals, defines candidate sensitive features, and filters and determines verification features using the Pearson correlation coefficient. Then, it extracts sensitive features based on the vibration signals of each window, sets feature weights, and combines three process scenarios—thin-wall, high-hardness, and stable cutting—after normalization, generates a weighted comprehensive feature value. When the number of comprehensive feature values ​​reaches the number of consecutive calculation windows, a sliding sequence is constructed. Finally, based on the first and second state thresholds and the number of anomaly tolerances, the initial, middle, and late tool wear stages are determined. Combining the tool condition output from the previous round with the current process scenario, the feature weights are dynamically adjusted to improve the accuracy and robustness of monitoring results under different wear stages and different machining scenarios.

Description

Technical Field

[0001] This invention relates to the fields of machining and intelligent manufacturing technology, specifically to a method for monitoring the status of CNC machine tool cutting tools based on multi-window feature registers. Background Technology

[0002] Tool condition is a key factor affecting machining quality and production efficiency in the machining process. In manufacturing applications, tools wear as they cut, and wear is divided into initial wear, intermediate wear, and late tool wear stages. Each stage has a different degree of impact on machining quality and tool life.

[0003] In the machining stage where machine tools perform cutting motions and the tool interacts substantially with the workpiece material, existing tool condition monitoring technologies commonly combine multi-sensor fusion and deep learning methods for prediction and classification. However, these methods require large amounts of training data and complex computing resources, making them difficult to adapt to the monitoring requirements of different processes.

[0004] In contrast, low-cost, high-real-time tool condition monitoring can be achieved by extracting features from a single type of sensor and performing logical judgments. In this process, determining the feature extraction window and making robust tool condition decisions based on the judgment results become key issues. Furthermore, the determination of tool condition in actual machining should be closely related to the machine tool's process scenario: in thin-wall machining, high-hardness machining, and other process scenarios, due to large fluctuations in cutting loads, significant vibration and impact, or accelerated tool wear rates, the judgment logic needs to be specifically optimized to achieve robust tool condition determination under different process scenarios.

[0005] Therefore, there is an urgent need for a real-time tool status recognition method that uses signal feature extraction and logical judgment to adaptively adjust weights at different cutting positions and different process stages to meet actual industrial needs.

[0006] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide a method for monitoring the status of CNC machine tool tools based on multi-window feature registers, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A method for monitoring the tool status of CNC machine tools based on multi-window feature registers, comprising the following steps:

[0010] Step 1: Acquire the vibration signal of the tool in real time during the current monitoring period, divide the current monitoring period into several time windows at equal intervals, extract candidate sensitive features of the vibration signal in each time window, calculate the correlation between each candidate sensitive feature and the tool wear width using the Pearson correlation coefficient, and then screen the sensitive features.

[0011] Step 2: Based on the tool wear stage under the previous monitoring time period, determine the feature weight of each sensitive feature under the current monitoring time period, generate the initial comprehensive feature value under each time window using a weighted fusion formula, and correct it to the comprehensive feature value according to the adjustment coefficient set by the process scenario. The process scenario is thin-walled, high-hardness or stable cutting machining.

[0012] Step 3: Set the first state threshold and the second state threshold. Construct a comprehensive feature value sequence from the comprehensive feature values ​​corresponding to each time window within the current monitoring period. In the comprehensive feature value sequence, count the number of comprehensive feature values ​​falling between the first and second state thresholds and the number exceeding the second state threshold. Determine the tool wear stage of the current monitoring period based on the comparison results with the preset abnormal tolerance number.

[0013] Furthermore, the candidate sensitive features include the peak value, mean, variance, root mean square, skewness, and kurtosis of the vibration signal.

[0014] Furthermore, the method for calculating the correlation between each candidate sensitive feature and the tool wear width, and then selecting sensitive features, is as follows:

[0015] The range of the flank wear width is set to 0mm-0.4mm, and 9 gradients are divided at 0.05mm intervals, with each gradient corresponding to a tool sample;

[0016] Under the same cutting parameters and stable cutting process, each tool sample is tested to obtain the vibration signal of each tool sample at the same time length. Then, for each tool sample, its corresponding candidate sensitive features are extracted. All kinds of candidate sensitive features of all tool samples are summarized to construct 6 candidate sensitive feature sequences.

[0017] The tool wear width values ​​of all tool samples are summarized to construct a VB value sequence. Based on the VB value sequence and each candidate sensitive feature sequence, the Pearson correlation coefficient between each candidate sensitive feature and the tool wear width value is calculated. If the absolute value of the Pearson correlation coefficient between a candidate sensitive feature and the tool wear width value is not less than a preset threshold, then this candidate sensitive feature is taken as a sensitive feature. This determines the type of sensitive feature and thus determines the sensitive features of the tool in each time window within the current monitoring period.

[0018] Furthermore, the method for generating the initial comprehensive feature values ​​for each time window using a weighted fusion approach is as follows:

[0019] The tool wear stage with a flank wear width of no more than 0.1 mm is defined as the initial tool wear stage; the tool wear stage with a flank wear width of more than 0.1 mm but no more than 0.25 mm is defined as the intermediate tool wear stage; and the tool wear stage with a flank wear width of more than 0.25 mm is defined as the late tool wear stage.

[0020] The previous monitoring period is defined as the monitoring period one time window forward along the time axis relative to the current monitoring period, and each individual monitoring period always maintains a continuous sequence. The fixed length of each time window means that adjacent monitoring time periods exist. There are 1 overlapping time window, differing by 1 non-overlapping time window. It is an integer not less than 10;

[0021] Based on the tool wear stage in the previous monitoring period, this is the [number]th [stage] in the current monitoring period. Each sensitive feature within a time window is assigned a corresponding feature weight. The specific process for setting the feature weights is as follows:

[0022] For any tool wear stage, extract the sensitive features and tool wear width of all tool samples under that tool wear stage. Calculate the absolute value of the Pearson correlation coefficient between each sensitive feature and the tool wear width under that tool wear stage, and use it as the initial weight reference value for the corresponding sensitive feature under that tool wear stage. With the cumulative value equal to 1 as a constraint, scale the initial weight reference values ​​of all sensitive features under that tool wear stage proportionally to obtain the feature weights corresponding to each sensitive feature under that tool wear stage.

[0023] For each time window within the current monitoring period, based on the tool wear stage determined in the previous monitoring period, the feature weights of each sensitive feature within that time window are determined. The sensitive features within that time window are then weighted and summed with their corresponding feature weights to obtain the initial comprehensive feature value for that time window. Finally, the feature value for the current monitoring period is obtained. The initial comprehensive feature values ​​for each time window.

[0024] Furthermore, for the three process scenarios of thin-wall machining, high-hardness machining, and stable cutting machining, independent adjustment coefficients are set for each process scenario, and the adjustment coefficients of the three process scenarios increase sequentially to determine the process scenario in which the tool is located in each time window. For each time window, the product of the initial comprehensive feature value and the corresponding adjustment coefficient is used as the corrected comprehensive feature value for that time window.

[0025] Furthermore, when monitoring was first initiated, the number of time windows did not reach [a certain threshold]. The following methods are used to calculate the initial and corrected comprehensive eigenvalues:

[0026] The initial tool wear stage is determined based on the back face wear width. The feature weights corresponding to each sensitive feature are obtained. Starting from the first acquisition of vibration signal, time windows are generated one by one and the sensitive features in each time window are extracted. For each newly added time window, based on the feature weights corresponding to the initial tool wear stage, the sensitive features in the window are weighted and summed with the corresponding feature weights to obtain the initial comprehensive feature value of the window. Then, the initial comprehensive feature value of each time window is multiplied by the adjustment coefficient of the corresponding process scenario to obtain the corrected comprehensive feature value of the window.

[0027] Until the cumulative generation The comprehensive feature value of each time window is determined based on the tool wear stage in the previous monitoring time period. The feature weight of each sensitive feature in the current monitoring time period is determined. The initial comprehensive feature value in each time window is generated by a weighted fusion formula and then corrected to a comprehensive feature value according to the adjustment coefficient set by the process scenario.

[0028] Furthermore, the first state threshold is less than the second state threshold, and both are greater than 0.

[0029] Furthermore, the method for determining the tool wear stage during the monitoring period based on the comparison results with the preset abnormal tolerance number is as follows:

[0030] A preset anomaly tolerance value is set, which is no greater than the number of time windows within the monitoring period and is not zero. For the comprehensive feature values ​​corresponding to each time window within the current monitoring period, they are constructed into a comprehensive feature value sequence in chronological order. The number of each comprehensive feature value within the sequence is counted, and the tool wear stage is determined by combining this with the anomaly tolerance value. The specific process is as follows:

[0031] When the number of comprehensive feature values ​​not less than the first state threshold in the comprehensive feature value sequence is less than the number of abnormal tolerance values, the corresponding tool wear stage is output as the initial tool wear stage.

[0032] When the number of comprehensive feature values ​​falling between the first and second state thresholds is greater than the number of abnormal tolerances, and the number of comprehensive feature values ​​not less than the second state threshold is less than the number of abnormal tolerances, the corresponding tool state is output as the intermediate tool wear stage.

[0033] When the number of comprehensive feature values ​​not less than the second state threshold is greater than the number of abnormal tolerance values, the corresponding tool wear stage output is the later tool wear stage.

[0034] Compared with the prior art, the beneficial effects of the present invention are:

[0035] This invention selects the most relevant sensitive features to different tool states from candidate sensitive features obtained from vibration signals based on the Pearson correlation coefficient. By setting weights for the sensitive features and adjusting the weights according to the process scenario and the tool state output from the previous comprehensive feature value sequence, multiple individual comprehensive feature values ​​are generated. This allows for rapid and real-time judgment of wear conditions for various tool materials, solving the problem of insufficient robustness in tool state judgment under complex working conditions without relying on deep learning. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the overall method flow of the present invention;

[0037] Figure 2 This is a statistical chart of the tool wear stages of the present invention;

[0038] Figure 3 This is a flowchart of the sensitive feature screening process of the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0040] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0041] Example:

[0042] Please see Figures 1 to 3 The present invention provides a technical solution:

[0043] A method for monitoring the tool status of CNC machine tools based on multi-window feature registers, comprising the following steps:

[0044] Step 1: Acquire the vibration signal of the tool in real time during the current monitoring period, divide the current monitoring period into several time windows at equal intervals, extract candidate sensitive features of the vibration signal in each time window, calculate the correlation between each candidate sensitive feature and the tool wear width using the Pearson correlation coefficient, and then screen the sensitive features.

[0045] First, the current monitoring time period is defined as an interval that covers a fixed number of time windows, starting from the current monitoring time and going backwards. The number of time windows must be no less than 10. Vibration signals from the CNC machine tool cutting process are continuously collected. Within the current monitoring time period, the vibration signals are divided into a fixed number of time windows at equal intervals. At the same time, at the beginning of each new monitoring time period, the earliest time window in the current monitoring time period is removed, and a newly generated latest time window is included to form a new current monitoring time period.

[0046] Candidate sensitive features of the vibration signal within each time window are extracted. These candidate sensitive features include the peak value, mean, variance, root mean square, skewness, and kurtosis of the vibration signal, such as... Figure 2 As shown, in the study of the change of flank wear width with cutting time, the growth rate of flank wear width (VB) is used as the core criterion for dividing the wear stages. 0-0.1VB / mm is considered initial wear, 0.1-0.25VB / mm is considered normal wear, and greater than 0.25VB / mm is considered severe wear. Therefore, the tool condition with a flank wear width of no more than 0.1mm is defined as the initial tool wear stage, the tool condition with a flank wear width greater than 0.1mm but not greater than 0.25mm is defined as the intermediate tool wear stage, and the tool condition with a flank wear width greater than 0.25mm is defined as the late tool wear stage. It is believed that when the tool condition is defined as the late tool wear stage, the tool needs to be replaced.

[0047] Among them, the peak value reflects the maximum amplitude of the vibration signal. After tool wear, the contact stability between the cutting edge and the workpiece decreases, leading to a sharp increase in the instantaneous vibration amplitude. The peak value can directly characterize this type of abnormal impact. The mean value reflects the overall energy level of the vibration signal. Increased tool wear will cause a continuous increase in cutting resistance, leading to an increase in the average energy of the vibration signal. The mean value can reflect the gradual change in wear. The variance reflects the dispersion of the vibration signal. In the initial stage of tool wear, the cutting state fluctuates little, and the variance value is stable. In the middle stage of tool wear, the cutting state begins to fluctuate irregularly, and the variance will show a significant increase. The root mean square (RMS) is used to measure the vibration signal. The effective value of the vibration signal can comprehensively characterize the energy intensity of the signal. Compared with the mean value, it can better highlight the contribution of large-amplitude vibrations and adapt to the changes in energy concentration caused by tool wear. Skewness reflects the asymmetry of the probability distribution of the vibration signal. Under normal cutting conditions, the vibration signal distribution is approximately symmetrical. When the tool suffers damage such as local chipping, the signal distribution will shift towards the large-amplitude direction, and the skewness value will deviate significantly. Kurtosis reflects the impact characteristics of the vibration signal and is extremely sensitive to the pulse component in the signal. Early micro-wear of the tool will cause high-frequency micro-impacts. Kurtosis can capture such weak anomalies in the early stage of wear and has early warning capability.

[0048] Typically, the wear width VB of the tool face is used as the dulling standard, as it has a significant impact on the quality of the machined surface. The Pearson correlation coefficient can accurately characterize the linear correlation strength between two variables. Using this method in the preliminary analysis stage, the correlation coefficient of the tool face wear width and the degree of correlation between candidate sensitive features is obtained. The specific process is as follows:

[0049] The range of flank wear width was set to 0mm-0.4mm. This is because during tool cutting, the flank face will inevitably contact the machined surface of the workpiece. Studies on the change of flank wear width with cutting time indicate that a higher flank wear width leads to deterioration of the machined surface roughness and dimensional inaccuracies. During the tool wear process, 0-0.1VB / mm is defined as initial wear, 0.1-0.25VB / mm as normal wear, and over 0.25VB / mm as severe wear. To ensure consistent sample gradient numbers for each tool wear stage, the upper limit of the gradient for the severe tool wear stage was set to 0.4mm, divided into 9 gradients at 0.05mm intervals. Each gradient corresponds to one tool sample, eliminating the interference of tool differences on the experiment. Each tool sample was tested under the same cutting parameters and stable cutting conditions to obtain vibration signals for each tool sample at the same time length. Then, for each tool sample, its corresponding candidate sensitive features were extracted. All candidate sensitive features of all tool samples were summarized to construct 6 candidate sensitive feature sequences, such as... Figure 3As shown, under nine gradients during the tool wear process, the peak value, mean, variance, root mean square, skewness, and kurtosis of the vibration signal are obtained respectively. By constructing six candidate sensitive feature sequences, the sensitive features can be screened by calculating the Pearson correlation coefficient. Under different tool models and different cutting materials, the response sensitivity of various candidate sensitive features of the vibration signal to the tool wear width is significantly different. Therefore, the tool wear width values ​​of all tool samples are summarized to construct a VB value sequence. Based on the VB value sequence and each candidate sensitive feature sequence, the Pearson correlation coefficient between each candidate sensitive feature and the tool wear width value is calculated. If the absolute value of the Pearson correlation coefficient between a candidate sensitive feature and the tool wear width value is not less than a preset threshold, then this candidate sensitive feature is regarded as a sensitive feature. The preset threshold can be 0.8. If the Pearson correlation coefficient calculated for skewness is 0.78, then skewness is considered not to be a sensitive feature.

[0050] Step 2: Based on the tool wear stage under the previous monitoring time period, determine the feature weight of each sensitive feature under the current monitoring time period, generate the initial comprehensive feature value under each time window using a weighted fusion formula, and correct it to the comprehensive feature value according to the adjustment coefficient set by the process scenario. The process scenario is thin-walled, high-hardness or stable cutting machining.

[0051] To achieve real-time monitoring and rapid anomaly identification of tool wear, the previous monitoring time period is set to be one time window forward along the time axis relative to the current monitoring time period, and each monitoring time period is always kept continuous. The fixed length of each time window means that adjacent monitoring time periods exist. There are 1 overlapping time window, differing by 1 non-overlapping time window. It must be an integer not less than 10, meaning that each time the monitoring time period is updated, only the most recently generated time window needs to be included and the earliest time window needs to be removed;

[0052] The sensitivity of sensitive features varies across different tool wear stages. For example, the study found that the peak value and kurtosis of vibration signals are more sensitive in the early stages of wear, while the mean and variance are less sensitive. Therefore, a feature weight is preset for each sensitive feature based on the current state of the tool to avoid diluting the contribution of highly sensitive features and amplifying the interference of low-sensitive features. Based on the tool wear stage in the previous monitoring period, the weight is set as the [number]th [stage] in the current monitoring period. Each sensitive feature within a time window is assigned a corresponding feature weight. The specific process for setting the feature weights is as follows:

[0053] Weighting is performed for the initial, intermediate, and late tool wear stages. For each tool wear stage, the corresponding data of each sensitive feature in all tool samples at that stage are first organized into a first sensitive feature sequence. At the same time, the tool face wear width data of all tool samples at that stage are collected to construct a corresponding first tool face wear width sequence. The absolute value of the Pearson correlation coefficient between each first sensitive feature sequence and the first tool face wear width sequence at that stage is calculated and used as the initial weight reference value of the corresponding sensitive feature at that stage. With the sum of the initial weight reference values ​​of all sensitive features equal to 1 as a constraint, the initial weight reference values ​​of each sensitive feature are adjusted proportionally to obtain the feature weights corresponding to each sensitive feature at each tool wear stage.

[0054] This method utilizes the Pearson correlation coefficient to quantify the inherent correlation between features and wear. For each time window within the current monitoring period, based on the tool wear stage determined in the previous monitoring period, the feature weights of each sensitive feature within that time window are determined. For example, if the tool wear stage output by a comprehensive feature value sequence is intermediate wear, the feature weights are directly replaced with the feature weights corresponding to the intermediate tool wear stage, serving as the feature weights for the latest time window generated in the next comprehensive feature value sequence. This ensures that the weights of sensitive features always adapt to the current tool wear stage, guaranteeing that the calculated comprehensive feature values ​​more accurately reflect the true tool wear state and preventing the underestimation of more sensitive features under different tool wear stages. The weighted sum of each sensitive feature within the time window and its corresponding feature weight is then used to obtain the initial comprehensive feature value for that time window, ultimately yielding the value for the current monitoring period. Initial comprehensive feature values ​​for each time window;

[0055] For each time window within the current monitoring period, based on the tool wear stage determined in the previous monitoring period, the feature weights of each sensitive feature within the time window are determined. The sensitive features within the time window are then weighted and summed with their corresponding feature weights to obtain the initial comprehensive feature value for that time window. Finally, the feature value for the current monitoring period is obtained. Initial comprehensive feature values ​​for each time window;

[0056] For three machining scenarios—thin-wall machining, high-hardness machining, and stable cutting machining—adjustment coefficients are set for each scenario. The adjustment coefficient for thin-wall machining is smaller than that for high-hardness machining, which is smaller than that for stable cutting machining. This offsets the inaccuracy of the comprehensive characteristic value caused by non-tool wear factors in different machining scenarios. After targeted adjustments, the comprehensive characteristic value more accurately reflects the tool condition rather than the vibration differences of the working condition itself. In the stable cutting machining scenario, the vibration signal is considered to reflect only the influence of tool wear itself. Therefore, the stable cutting machining scenario is used as the benchmark working condition, and the adjustment coefficient is set to 1. In high-hardness machining, the cutting resistance is large, the load fluctuation and vibration impact are strong, which will make the overall sensitive characteristic value higher. Therefore, the adjustment coefficient is less than 1. The core problem of thin-wall machining is that the cutting force causes elastic deformation of the workpiece. Combined with the vibration of the cutting itself, the signal amplitude is significantly increased, which is more obvious than that of high-hardness machining. Its adjustment coefficient is the smallest and less than 1. The product of the initial comprehensive characteristic value and the corresponding adjustment coefficient is used as the corrected comprehensive characteristic value under this time window.

[0057] When monitoring is first initiated, a complete monitoring period is not yet formed, meaning the number of time windows has not been reached. The following methods are used to calculate the initial and corrected comprehensive eigenvalues:

[0058] The initial tool wear stage is determined based on the flank wear width. Feature weights corresponding to each sensitive feature are obtained. Starting from the first vibration signal acquisition, time windows are generated sequentially, and sensitive features within each time window are extracted. For each new time window, based on the feature weights corresponding to the initial tool wear stage, each sensitive feature within the window is weighted and summed with its corresponding feature weight to obtain the initial comprehensive feature value for that window. Then, the initial comprehensive feature value for each time window is multiplied by the adjustment coefficient of the corresponding process scenario to obtain the corrected comprehensive feature value for that window. This process continues until a total of [number of windows] are generated. Based on the comprehensive feature values ​​of each time window, the feature weights of each sensitive feature in the current monitoring time period can be determined according to the tool wear stage in the previous monitoring time period. The initial comprehensive feature value in each time window is generated by a weighted fusion formula, and then corrected to a comprehensive feature value according to the adjustment coefficient set by the process scenario.

[0059] Step 3: Set the first state threshold and the second state threshold, construct a comprehensive feature value sequence from the comprehensive feature values ​​corresponding to each time window within the current monitoring period, count the number of comprehensive feature values ​​falling between the first and second state thresholds and the number exceeding the second state threshold in the comprehensive feature value sequence, and determine the tool wear stage of the current monitoring period based on the comparison results with the preset abnormal tolerance number.

[0060] Set a first state threshold and a second state threshold, where the first state threshold is less than the second state threshold, and both values ​​are greater than 0. The first state threshold serves as the boundary reference between initial and intermediate wear, while the second state threshold serves as the boundary reference between intermediate and late wear. This hierarchical division of the dual thresholds enables fine differentiation of tool states. The first state threshold value must satisfy the following condition: when the tool state is initially in the intermediate tool wear stage, the comprehensive feature value calculated based on the feature weight corresponding to the initial wear is not less than the first state threshold. Similarly, the second state threshold value must satisfy the following condition: when the tool state is initially in the late tool wear stage, the comprehensive feature value calculated based on the feature weight corresponding to the intermediate wear is not less than the second state threshold. For example, through offline experiments for calibration, the comprehensive feature values ​​calculated when the tool flank wear width is 0.1VB / mm and 0.25VB / mm are used as the first and second state thresholds, respectively.

[0061] A preset anomaly tolerance number is established. This anomaly tolerance number has a value range that is no greater than the number of divided time windows and is not zero. For the comprehensive feature values ​​corresponding to each time window within the current monitoring period, they are constructed into a comprehensive feature value sequence according to their chronological order. The number of each comprehensive feature value in the comprehensive feature value sequence is counted, and the tool status is determined in combination with the anomaly tolerance number. The specific process is as follows:

[0062] When the number of comprehensive feature values ​​not less than the first state threshold in the comprehensive feature value sequence is less than the number of abnormal tolerance values, the first state threshold is the critical reference between initial wear and intermediate wear. When only a few windows have comprehensive feature values ​​that reach or exceed the threshold, it indicates that the data exceeding the threshold is mostly transient interference and no intervention is required. The corresponding tool state is output as the initial tool wear stage.

[0063] When the number of comprehensive feature values ​​not less than the first state threshold is greater than the number of abnormal tolerance values, and the number of comprehensive feature values ​​not less than the second state threshold is less than the number of abnormal tolerance values, it indicates that the tool wear has exceeded the initial critical threshold. At the same time, the number of windows not less than the second state threshold has not reached the number of abnormal tolerance values, indicating that the tool has not yet shown the serious characteristics of later wear. Monitoring needs to be strengthened, but the tool does not need to be replaced immediately. Therefore, the corresponding tool status is output as the intermediate tool wear stage.

[0064] When the number of comprehensive feature values ​​not less than the second state threshold is greater than the number of abnormal tolerance values, it indicates that most time windows are greater than the critical feature value of the corresponding tool wear in the later stage, reflecting that the tool cutting accuracy will drop significantly, which may lead to problems such as workpiece scrapping and increased machine tool vibration. Therefore, the corresponding tool state is output as the later tool wear stage.

[0065] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0066] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0067] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0068] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for monitoring the condition of CNC machine tool tools based on multi-window feature registers, characterized in that, The specific steps include: Step 1: Acquire the vibration signal of the tool in real time during the current monitoring period, divide the current monitoring period into several time windows at equal intervals, extract candidate sensitive features of the vibration signal in each time window, calculate the correlation between each candidate sensitive feature and the tool wear width using the Pearson correlation coefficient, and then screen the sensitive features. Step 2: Based on the tool wear stage under the previous monitoring time period, determine the feature weight of each sensitive feature under the current monitoring time period, generate the initial comprehensive feature value under each time window using a weighted fusion formula, and correct it to the comprehensive feature value according to the adjustment coefficient set by the process scenario. The process scenario is thin-walled, high-hardness or stable cutting machining. Step 3: Set the first state threshold and the second state threshold, construct a comprehensive feature value sequence from the comprehensive feature values ​​corresponding to each time window within the current monitoring period, count the number of comprehensive feature values ​​falling between the first and second state thresholds and the number exceeding the second state threshold in the comprehensive feature value sequence, and determine the tool wear stage of the current monitoring period based on the comparison results with the preset abnormal tolerance number. The candidate sensitive features include the peak value, mean, variance, root mean square, skewness, and kurtosis of the vibration signal; The range of the flank wear width is set to 0mm-0.4mm, and 9 gradients are divided at 0.05mm intervals, with each gradient corresponding to a tool sample; Under the same cutting parameters and stable cutting process, each tool sample is tested to obtain the vibration signal of each tool sample at the same time length. Then, for each tool sample, its corresponding candidate sensitive features are extracted. All kinds of candidate sensitive features of all tool samples are summarized to construct 6 candidate sensitive feature sequences. The tool wear width values ​​of all tool samples are summarized to construct a VB value sequence. Based on the VB value sequence and each candidate sensitive feature sequence, the Pearson correlation coefficient between each candidate sensitive feature and the tool wear width value is calculated. If the absolute value of the Pearson correlation coefficient between a candidate sensitive feature and the tool wear width value is not less than a preset threshold, then this candidate sensitive feature is taken as a sensitive feature. This determines the type of sensitive feature and thus determines the sensitive features of the tool in each time window within the current monitoring period. The tool wear stage with a flank wear width of no more than 0.1 mm is defined as the initial tool wear stage; the tool wear stage with a flank wear width of more than 0.1 mm but no more than 0.25 mm is defined as the intermediate tool wear stage; and the tool wear stage with a flank wear width of more than 0.25 mm is defined as the late tool wear stage. The previous monitoring period is defined as the monitoring period one time window forward along the time axis relative to the current monitoring period, and each individual monitoring period always maintains a continuous sequence. The fixed length of each time window means that adjacent monitoring time periods exist. There are 1 overlapping time window, differing by 1 non-overlapping time window. It is an integer not less than 10; Based on the tool wear stage in the previous monitoring period, this is the [number]th [stage] in the current monitoring period. Each sensitive feature within a time window is assigned a corresponding feature weight. The specific process for setting the feature weights is as follows: For any tool wear stage, extract the sensitive features and tool wear width of all tool samples under that tool wear stage. Calculate the absolute value of the Pearson correlation coefficient between each sensitive feature and the tool wear width under that tool wear stage, and use it as the initial weight reference value for the corresponding sensitive feature under that tool wear stage. With the cumulative value equal to 1 as a constraint, scale the initial weight reference values ​​of all sensitive features under that tool wear stage proportionally to obtain the feature weights corresponding to each sensitive feature under that tool wear stage. For each time window within the current monitoring period, based on the tool wear stage determined in the previous monitoring period, the feature weights of each sensitive feature within that time window are determined. The sensitive features within that time window are then weighted and summed with their corresponding feature weights to obtain the initial comprehensive feature value for that time window. Finally, the feature value for the current monitoring period is obtained. Initial comprehensive feature values ​​for each time window; When monitoring is first started, the number of time windows has not yet been reached. The following methods are used to calculate the initial and corrected comprehensive eigenvalues: The initial tool wear stage is determined based on the back face wear width. The feature weights corresponding to each sensitive feature are obtained. Starting from the first acquisition of vibration signal, time windows are generated one by one and the sensitive features in each time window are extracted. For each newly added time window, based on the feature weights corresponding to the initial tool wear stage, the sensitive features in the window are weighted and summed with the corresponding feature weights to obtain the initial comprehensive feature value of the window. Then, the initial comprehensive feature value of each time window is multiplied by the adjustment coefficient of the corresponding process scenario to obtain the corrected comprehensive feature value of the window. Until the cumulative generation The comprehensive feature value of each time window is determined based on the tool wear stage in the previous monitoring time period. The feature weight of each sensitive feature in the current monitoring time period is determined. The initial comprehensive feature value of each time window is generated by a weighted fusion formula and then corrected to a comprehensive feature value according to the adjustment coefficient set by the process scenario.

2. The method for monitoring the status of CNC machine tool tools based on multi-window feature registers according to claim 1, characterized in that: For three process scenarios—thin-wall machining, high-hardness machining, and stable cutting machining—independent adjustment coefficients are set for each process scenario, and the adjustment coefficients for the three process scenarios increase sequentially. This determines the process scenario in which the tool is located in each time window. For each time window, the product of the initial comprehensive feature value and the corresponding adjustment coefficient is used as the corrected comprehensive feature value for that time window.

3. The method for monitoring the status of CNC machine tool tools based on multi-window feature registers according to claim 1, characterized in that: The threshold for the first state is less than the threshold for the second state, and both are greater than 0.

4. The method for monitoring the status of CNC machine tool tools based on multi-window feature registers according to claim 2, characterized in that: The method for determining the tool wear stage during the monitoring period based on the comparison results with the preset abnormal tolerance number is as follows: A preset anomaly tolerance value is set, which is no greater than the number of time windows within the monitoring period and is not zero. For the comprehensive feature values ​​corresponding to each time window within the current monitoring period, they are constructed into a comprehensive feature value sequence in chronological order. The number of each comprehensive feature value within the sequence is counted, and the tool wear stage is determined by combining this with the anomaly tolerance value. The specific process is as follows: When the number of comprehensive feature values ​​not less than the first state threshold in the comprehensive feature value sequence is less than the number of abnormal tolerance values, the corresponding tool wear stage is output as the initial tool wear stage. When the number of comprehensive feature values ​​falling between the first and second state thresholds is greater than the number of abnormal tolerances, and the number of comprehensive feature values ​​not less than the second state threshold is less than the number of abnormal tolerances, the corresponding tool state is output as the intermediate tool wear stage. When the number of comprehensive feature values ​​not less than the second state threshold is greater than the number of abnormal tolerance values, the corresponding tool wear stage output is the later tool wear stage.

Images