Cutter state monitoring method based on dynamic envelope line

Abstract

The invention provides a cutter state monitoring method based on a dynamic envelope, and belongs to the technical field of cutter wear detection, and the method comprises the following steps: S1, carrying out the preprocessing of an original three-phase current signal of a cutter through calculating a root mean square (RMS); s2, calculating statistical characteristics of the preprocessed current signal based on a 3 sigma rule, and extracting characteristics reflecting the current change trend of the cutter; s3, generating a dynamic envelope line for the generated statistical characteristics according to a learning mode, and taking the dynamic envelope line as a processing upper and lower threshold line; s4, entering a working mode after learning is completed, and comparing the real-time processing data with the upper and lower threshold lines to realize threshold exceeding alarm; and S5, a multi-level alarm pushing algorithm is adopted, and the false alarm rate is reduced. The method can monitor the machining state of the cutter in real time, discover abnormity in time and give an alarm, improve the machining quality and efficiency and reduce the equipment damage risk.

Description

Technical Field

[0001] This invention belongs to the field of tool wear detection technology, and particularly relates to a tool condition monitoring method based on dynamic envelope. Background Technology

[0002] As the manufacturing industry transforms and upgrades towards automation and intelligence, automated production lines consisting of multiple CNC machine tools have become the mainstream production mode. While this mode improves processing efficiency and reduces labor costs, it also places higher demands on the continuity and reliability of the production process. The condition of the process system, especially the cutting tools, directly affects the processing efficiency, product quality, and equipment operation safety of the production line. In highly automated production lines, only a small number of personnel are typically on duty. If a cutting tool on a machine tool suddenly breaks or wears excessively and is not detected in time, it can easily lead to the scrapping of a batch of workpieces, damage to the machine tool spindle, or even production line shutdown, resulting in significant economic losses. Therefore, researching efficient and reliable online monitoring technology for tool status to achieve real-time early warning of tool anomalies is a key engineering problem that urgently needs to be solved to ensure the stable and efficient operation of automated production lines.

[0003] Currently, mainstream tool condition monitoring methods can be divided into two main categories: direct measurement methods and indirect measurement methods. Direct measurement methods (such as machine vision and laser measurement) have high accuracy, but are limited by factors such as the machining environment, coolant, and chip obstruction, making it difficult to implement online applications. Indirect measurement methods indirectly determine the tool condition by analyzing physical signals related to the tool condition, mainly including: (1) Cutting force monitoring: The signal directly reflects the cutting process, but requires the installation of expensive force measuring instruments, resulting in high engineering application costs; (2) Vibration signal analysis: Sensitive to early wear, but the sensor installation position is demanding and easily affected by environmental noise; (3) Acoustic emission technology: Extremely sensitive to damage, but signal processing is complex, and the sensor lifespan and reliability face challenges under harsh working conditions; (4) Motor current / power monitoring: Utilizing the current / power signals of the machine tool spindle or feed motor, since its sensor is an inherent configuration of the machine tool itself, no additional hardware is required, making it one of the methods with the greatest potential for engineering applications. However, existing current / power-based monitoring methods mostly rely on setting fixed empirical thresholds or require the establishment of a benchmark model under specific stable working conditions. However, in actual production, factors such as changes in processing parameters, fluctuations in workpiece material, differences in the initial performance of cutting tools, and time-varying load characteristics of machine tools make fixed thresholds or static models poorly adaptable, prone to false alarms or missed alarms, and difficult to achieve robust and accurate monitoring. Summary of the Invention

[0004] To address the aforementioned problems, this invention proposes a tool condition monitoring method based on a dynamic envelope. The core of this method lies in its approach: instead of using a fixed threshold, it learns from tool current signals during historical normal machining processes to automatically generate a dynamic envelope that adapts to specific tool, workpiece, and process parameters. This dynamic envelope effectively characterizes the reasonable fluctuation range of the current signal during normal machining. In the real-time tool monitoring phase, the system compares the collected real-time tool current signal with an alarm threshold. Through threshold exceedance judgment and a multi-level alarm mechanism, it achieves accurate, real-time, and reliable early warning of abnormal tool conditions, effectively overcoming the shortcomings of traditional fixed threshold methods in terms of adaptability and improving the intelligence level and engineering practical value of the monitoring system.

[0005] Its characteristic is that it includes the following process:

[0006] S1, Preprocessing and feature extraction of the raw current signal of the tool;

[0007] S2, the learning mode generates a dynamic envelope based on multiple normal processing data;

[0008] S3 compares real-time machining data with the dynamic envelope to determine the tool status and trigger an over-threshold alarm.

[0009] S4, a multi-level alarm push algorithm.

[0010] S1 includes:

[0011] S1.1, calculate the RMS of the tool current signal;

[0012] During normal operation, the input is the current signal acquired by the machine tool cutting tool. The current signal has three phases: U, V, and W. The RMS is calculated for each of the three-phase input signals, and the formula is as follows:

[0013] Formula (1);

[0014] in, , , For a three-phase current signal, the effective value, RMS, which reflects the current trend, is obtained by calculating the root mean square of the current values ​​in the U, V, and W phases.

[0015] S1.2: Based on the RMS value obtained in S1.1, the statistical characteristics of each segment are calculated using the 3σ rule of the normal distribution through a sliding window.

[0016] Let the RMS dataset be... ], through 3 The statistical characteristics obtained by the rule-based sliding window calculation are: ], of which 3 The formula for calculating the rule is as follows:

[0017] Formula (2);

[0018] in, The RMS value obtained from formula (1);

[0019] This reflects the central trend of the data, i.e., the baseline level under normal processing conditions;

[0020] This reflects the degree of dispersion of the reaction data, that is, the normal fluctuation range during the processing.

[0021] The statistical characteristics calculated based on formula (2) The input data obtained from multiple normal processing steps is called the learning data and serves as the learning sample set.

[0022] S2 includes:

[0023] S2.1: Filtering and cleaning the learning data.

[0024] First, the length of the learning data is calculated based on the learning sample set obtained from S1, and its maximum value is taken.

[0025] Secondly, 50% of this maximum value is set as the effective length threshold of the learning sample set;

[0026] Finally, the learning sample set is filtered based on this length threshold, and only sample data with sequence length not less than the length threshold are retained as learning data to remove abnormal samples.

[0027] S2.2: Construct segmented intervals and extract corresponding data.

[0028] The filtered learning data of S2.1 are organized into a set of learning curves, let the set of learning curves be . , ;

[0029] in, Here, N represents the number of learning curves;

[0030] The set of lengths for each curve is as follows: ,

[0031] First, calculate the length statistics:

[0032] Formula (3);

[0033] Then, based on formula (3), the length values ​​are extracted and sorted, and the sets are divided into segments in ascending order. ;

[0034] in, For data in the segmented set, The number of segments, based on the set of segments. Extract the segments that fall within that set. Interval data ;

[0035] S2.3: To Perform calculations to generate the basic envelope;

[0036] First, regarding S2.2 The initial envelope is generated by taking the maximum and minimum values ​​point by point along the index direction;

[0037] Then, the peak and valley values ​​of the initial envelope are calculated, and the valley and peak regions are smoothed by Gaussian filtering to obtain the basic envelope.

[0038] Then, the peak and valley values ​​of the initial envelope are calculated, and the valley and peak regions are smoothed by Gaussian filtering to obtain a smoothed initial envelope.

[0039] Finally, the maximum value of the learning curve set C and the smooth envelope is taken point by point along the index direction to calibrate the initial envelope and obtain the basic envelope, ensuring that the basic envelope can cover all learning curves.

[0040] The Gaussian filtering formula is:

[0041] Formula (4);

[0042] in, The initial envelope value, The standard deviation is denoted as .

[0043] S2.4: Inflection point detection and neighborhood replacement generate dynamic envelope;

[0044] First, peak points in the effective current value are identified based on preset peak significance, minimum spacing, and height thresholds. These peaks are then sorted according to their significance, and a specified number of significant peaks are selected as envelope inflection points. To obtain the inflection point location.

[0045] Secondly, peak replacement is performed on the data in the neighborhood of the inflection point to generate a dynamic envelope with relaxed characteristics. That is, for each inflection point The system determines the replacement intervals on the left and right sides of a preset neighborhood range, sorts the inflection points according to their values, and replaces the data in the neighborhood corresponding to each inflection point with the amplitude of that inflection point in sequence. The specific implementation is as follows:

[0046] Let the basic envelope be ,in Let M be the length of the basic envelope, and let the set of the detected M inflection point locations be denoted as . The corresponding dynamic envelope amplitude with relaxed characteristics ;

[0047] The set of inflection point locations is All inflection points are sorted according to their corresponding amplitudes to obtain an ordered sequence:

[0048] Formula (5);

[0049] in For inflection point detection index, For the corresponding amplitude, and ;

[0050] Preset neighborhood radius For each ordered inflection point Its neighborhood coverage range is defined as:

[0051] Formula (6);

[0052] Then the loose envelope in step S2.4 This is achieved through the following substitution process:

[0053] If and only if Formula (7);

[0054] And for any If it belongs to multiple Then the largest in the sorted order corresponding As the final value. Not subject to any Covered Keep the original value .

[0055] Finally, the dynamic envelope is obtained by replacing the regions corresponding to the basic envelope with multiple horizontal inflection point regions.

[0056] S3 includes:

[0057] The real-time machining curve of the tool is compared with the dynamic envelope of the corresponding step S2.4. When the real-time machining curve exceeds the dynamic envelope, the system generates an alarm, thereby realizing the alarm of the tool's real-time status exceeding the threshold.

[0058] The real-time machining curve is obtained by accumulating the effective values ​​of the current signals collected by the machine tool during normal machining.

[0059] S4 includes:

[0060] When the real-time data of the current signal triggers an over-threshold alarm, the degree of deviation of the alarm data from the set alarm push threshold is compared. When the degree of deviation is higher than the alarm push threshold, an alarm is pushed based on step S3.

[0061] When the real-time data of the current signal triggers an over-threshold alarm but does not exceed the alarm push threshold, the number of consecutive alarms is accumulated. When the alarm push threshold is reached, an alarm is pushed.

[0062] If the algorithm does not meet the above two conditions, no alarm will be pushed.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] This invention discloses a tool status monitoring method based on a dynamic envelope. The system collects current data multiple times during normal tool machining, uses a segmented processing strategy combined with Gaussian filtering to generate a basic envelope, and further optimizes the basic envelope to a dynamic envelope with relaxed characteristics through inflection point detection and peak (valley) replacement, thus solving the problem of false alarms caused by data deviation. Simultaneously, a multi-level alarm mechanism is incorporated, considering not only single data exceedances but also identifying continuous minor anomalies. Furthermore, the system supports adaptive learning, continuously optimizing the reference standard based on normal machining data. This system can monitor the tool machining status in real time, promptly detect anomalies and issue alarms, improving machining quality and efficiency while reducing the risk of equipment damage. Attached Figure Description

[0065] Figure 1 This is a flowchart illustrating the overall process of tool monitoring in this invention.

[0066] Figure 2 This is the original current data for this invention.

[0067] Figure 3 The learning curve and basic envelope of this invention.

[0068] Figure 4 This is the dynamic envelope of the present invention.

[0069] Figure 5 This is the alarm curve for the working mode of the present invention. Detailed Implementation

[0070] The present invention will be further described below with reference to embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0071] like Figure 1-5The overall flowchart of the tool condition monitoring method based on dynamic envelope provided by this invention is as follows: Figure 1 As shown, the specific steps are as follows:

[0072] Step 1: Preprocessing of tool current signal. The original three-phase current signal is a periodic alternating signal. Its instantaneous value fluctuates with time and contains a large amount of high-frequency noise. Directly using the instantaneous value cannot accurately reflect the actual effect of the signal.

[0073] By combining the three-phase currents to calculate a single RMS, a single scalar value reflecting the overall current intensity of the three-phase system can be obtained, transforming the dynamic, time-varying instantaneous current signal into an equivalent static index characterizing the signal's capability. By continuously tracking the changing trend of this RMS value, faults such as abnormal tool wear, chipping, or breakage can be effectively identified.

[0074] In this example, a packet of current data is collected every 250ms. Each packet of current data contains 50 U, V, and W three-phase current values. An RMS value is calculated for each U, V, and W three-phase current value, resulting in 50 RMS values.

[0075] After accumulating 200 RMS values, the 3σ rule of normal distribution is used to calculate the statistical characteristics of each data segment through a sliding window, thereby amplifying the data when tool failure occurs, increasing the sensitivity of the failure, and reducing the amount of data. The window size is set to 5, and a statistical characteristic value (i.e., the mean plus three standard deviations) is calculated for every 5 RMS values ​​of the 200 RMS values, resulting in 40 valid values.

[0076] Step 2: The learning mode generates a dynamic envelope based on multiple normal processing data.

[0077] S2.1: By continuously accumulating the effective values ​​from step 1, three real-time machining curves are generated during normal tool processing, i.e., learning process curves. Each curve represents the effective values ​​collected from the start to the end of machining for a workpiece. The maximum length of the three learning curves is calculated. If the length of one of the learning curves is less than 50% of the maximum length, that curve is removed, and only the curves with the effective length are retained for learning.

[0078] S2.2: Alignment ( After indexing the learning curves, the lengths of the effective curves are first calculated and sorted in ascending order. Then, using the sorted curve lengths as the dividing points, the interval from 0 to the shortest curve length is successively selected as the first segmented interval, the interval from the shortest curve length to the second shortest curve length is selected as the second segmented interval, and so on, until all curve data are covered, finally forming a set of segmented intervals.

[0079] S2.3: An example of the calculation result for the first segmented interval is as follows:

[0080]

[0081] First, for each segmented interval, find the maximum value point by point along the index direction. and minimum value This yields the set of maximum and minimum values ​​for n learning curves across different segment intervals. The calculation results for the first segment interval are as follows:

[0082]

[0083]

[0084] Secondly, Gaussian filtering is then applied to each... , The valley and peak regions are smoothed out.

[0085] ;

[0086] ;

[0087] Finally, along the index direction, the maximum value of the original learning curve set and the initial envelope is taken point by point, and all smoothed curves are then processed. , The basic envelope is obtained by concatenating the indices; the upper basic envelope is then... The array obtained by concatenation The lower basic envelope is The array obtained by concatenation .

[0088] ;

[0089] ;

[0090] S2.4: The version generated from S2.3 , Inflection point detection is performed to generate a dynamic envelope.

[0091] First, find , The peak and trough points are defined, with a peak (trough) significance of 100. When the difference between two adjacent valid values ​​is greater than 100, they are judged as peak / trough values. The peaks (trough values) are sorted, and the top 30 peaks (trough values) are taken as inflection points.

[0092] Secondly, replace the data adjacent to the inflection point with a replacement range of 300. Find the peak (valley) point and replace it in ascending order to ensure that the large peak is not covered by the small peak. Sort the valleys in descending order to ensure that the small valleys are not covered by the large valleys.

[0093] Finally, the dynamic envelope is obtained by replacing the regions corresponding to the basic envelope with multiple horizontal inflection point regions.

[0094] Step 4: Compare the real-time machining data of the tool with the dynamic envelope. When the real-time machining data exceeds the dynamic envelope, the system generates an alarm, thereby realizing the alarm of the tool's real-time status exceeding the threshold.

[0095] Step 5: Alarm Push Logic. When real-time data triggers an over-threshold alarm, the algorithm calculates the deviation of the alarm data from the alarm threshold. If the deviation exceeds twice the dynamic envelope (alarm push threshold), the algorithm immediately pushes an alarm. When real-time data triggers an over-threshold alarm but does not exceed the alarm push threshold, the algorithm accumulates the number of consecutive alarms. When the number of alarms reaches 5, the algorithm pushes an alarm. If the algorithm does not meet either of the above two conditions, no alarm is pushed.

[0096] The above description is merely a preferred embodiment of this application and is not intended to limit 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 protection scope of this application.

[0097] While the specific embodiments of the present invention have been described above, they are not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A tool condition monitoring method based on dynamic envelope, characterized in that, Includes the following processes: S1, Preprocessing and feature extraction of the raw current signal of the tool; S2, the learning mode generates a dynamic envelope based on multiple normal processing data; S3 compares real-time machining data with a dynamic envelope to determine the tool status and trigger an over-threshold alarm. S4, a multi-level alarm push algorithm.

2. The tool condition monitoring method based on dynamic envelope according to claim 1, characterized in that, In step S1, the following steps are performed; S1.1, calculate the RMS of the tool current signal; During normal operation, the input is the current signal acquired by the machine tool cutting tool. The current signal has three phases: U, V, and W. The RMS is calculated for each of the three-phase input signals, and the formula is as follows: Official (1); in, , , For a three-phase current signal, the effective value, RMS, which reflects the current trend, is obtained by calculating the root mean square of the current values ​​in the U, V, and W phases. S1.2: Based on the RMS value obtained in S1.1, the statistical characteristics of each segment are calculated using the 3σ rule of the normal distribution through a sliding window. Let the RMS dataset be... ], through 3 The statistical characteristics obtained by the rule-based sliding window calculation are: ], of which 3 The formula for calculating the rule is as follows: Official (2); in, The RMS value obtained from formula (1); This reflects the central trend of the data, i.e., the baseline level under normal processing conditions; This reflects the degree of dispersion of the data, i.e., the normal fluctuation range during the processing. The statistical characteristics calculated based on formula (2) The input data obtained from multiple normal processing steps is called the learning data and serves as the learning sample set.

3. The tool condition monitoring method based on dynamic envelope according to claim 1, characterized in that, In step S2, the following steps are performed; S2.1: Filtering and cleaning the learning data; First, the length of the learning data is calculated based on the learning sample set obtained from S1, and its maximum value is taken. Secondly, 50% of this maximum value is set as the effective length threshold of the learning sample set; Finally, the learning sample set is filtered based on this length threshold, and only sample data with sequence length not less than the length threshold are retained as learning data to remove abnormal samples. S2.2: Construct segmented intervals and extract corresponding data; The filtered learning data of S2.1 are organized into a set of learning curves, let the set of learning curves be . , ; in, Here, N represents the number of learning curves; The set of lengths for each curve is as follows: , First, calculate the length statistics: Official (3); Then, based on formula (3), the length values ​​are extracted and sorted, and the sets are divided into segments in ascending order. ; in, For data in the segmented set, The number of segments, based on the set of segments. Extract the segments that fall within that set. Interval data ; S2.3: To Perform calculations to generate the basic envelope; First, regarding S2.2 The initial envelope is generated by taking the maximum and minimum values ​​point by point along the index direction; Then, the peak and valley values ​​of the initial envelope are calculated, and the valley and peak regions are smoothed by Gaussian filtering to obtain the basic envelope. Secondly, the peak and valley values ​​of the initial envelope are calculated, and the valley and peak regions are smoothed by Gaussian filtering to obtain a smoothed initial envelope. Finally, the maximum value of the learning curve set C and the smooth envelope is taken point by point along the index direction to calibrate the initial envelope and obtain the basic envelope, ensuring that the basic envelope can cover all learning curves. The Gaussian filtering formula is: Official (4); in, The initial envelope value, Standard deviation; S2.4: Inflection point detection and neighborhood replacement generate dynamic envelope; First, peak points in the effective current value are identified based on preset peak significance, minimum spacing, and height thresholds. These peaks are then sorted according to their significance, and a specified number of significant peaks are selected as envelope inflection points. Obtain the inflection point location; Secondly, peak replacement is performed on the data in the neighborhood of the inflection point to generate a dynamic envelope with relaxed characteristics. That is, for each inflection point The system determines the replacement intervals on the left and right sides of a preset neighborhood range, sorts the inflection points according to their values, and replaces the data in the neighborhood corresponding to each inflection point with the amplitude of that inflection point in sequence. The specific implementation is as follows: Let the basic envelope be ,in Let M be the length of the basic envelope, and let the set of the detected M inflection point locations be denoted as . The corresponding dynamic envelope amplitude with relaxed characteristics ; The set of inflection point locations is All inflection points are sorted according to their corresponding amplitudes to obtain an ordered sequence: Official (5); in For inflection point detection index, For the corresponding amplitude, and ; Preset neighborhood radius For each ordered inflection point Its neighborhood coverage range is defined as: Official (6); Then the loose envelope in step S2.4 This is achieved through the following substitution process: If and only if Formula (7); And for any If it belongs to multiple Then the largest in the sorted order corresponding As the final value. Not subject to any Covered Keep the original value ; Finally, the dynamic envelope is obtained by replacing the regions corresponding to the basic envelope with multiple horizontal inflection point regions.

4. The tool condition monitoring method based on dynamic envelope according to claim 1, characterized in that, In step S3, The real-time machining curve of the tool is compared with the dynamic envelope of the corresponding step S2.

4. When the real-time machining curve exceeds the dynamic envelope, the system generates an alarm, thereby realizing the alarm of the tool's real-time status exceeding the threshold. The real-time machining curve is obtained by accumulating the effective values ​​of the current signals collected by the machine tool during normal machining.

5. The tool condition monitoring method based on dynamic envelope according to claim 1, characterized in that, In step S4, the alarm push logic is implemented; When the real-time data of the current signal triggers an over-threshold alarm, the degree of deviation of the alarm data from the set alarm push threshold is compared. When the degree of deviation is higher than the alarm push threshold, an alarm is pushed based on step S3. When the real-time data of the current signal triggers an over-threshold alarm but does not exceed the alarm push threshold, the number of consecutive alarms is accumulated. When the alarm push threshold is reached, an alarm is pushed. If the algorithm does not meet the above two conditions, no alarm will be pushed.

Images