Intelligent analysis method for operation data of numerical control machine tool

Abstract

The present application relates to the technical field of electric digital data processing, in particular to a kind of numerical control machine operation data intelligent analysis method.The present application first obtains numerical control machine operation data, establishes sample space and obtains the abnormal score of data point;Further in combination with the distribution characteristics of data point in sample space, the abnormal score of data point is corrected, and the corrected abnormal score of each data point is obtained;Further analysis of the salient features of data point obtains the abnormal possibility of each data point;Finally, according to the abnormal possibility of each data point, abnormal data points are screened out.The present application is aimed at the problem that traditional isolated forest algorithm is not accurate for evaluating abnormal data with local clustering characteristics, by analyzing the distribution characteristics of data point, and combining the local and global salient features of the corrected abnormal score, more accurate abnormal evaluation is carried out on data points, and more accurate abnormal data points are screened out, to improve production efficiency and ensure product quality.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing technology, and specifically to an intelligent analysis method for CNC machine tool operation data. Background Technology

[0002] Machine tool production data refers to the various data generated by machine tools during the production process in the manufacturing industry. This data can cover multiple aspects, from the machine tool's operating status to quality control in the production process. Scientific monitoring and analysis of abnormal data generated during machine tool production can help identify key factors affecting product quality, thereby improving production efficiency, optimizing product quality, reducing operating costs, and enhancing the efficiency and flexibility of the supply chain.

[0003] Currently, when using isolated forests to detect anomalies in machine tool production data, the isolated forest method only focuses on the density of data in the sample space and does not reflect the location distribution characteristics of data points in the sample space. The distribution location of some data may also reflect its abnormal characteristics. Therefore, some locally dense abnormal production data may be misjudged as normal production data, thereby reducing the stability of CNC machine tool operation. Summary of the Invention

[0004] To address the technical problem of inaccurate analysis of CNC machine tool operation data using traditional isolated forest algorithms, this invention aims to provide an intelligent analysis method for CNC machine tool operation data. The specific technical solution adopted is as follows:

[0005] Acquire CNC machine tool operation data; establish a sample space based on the CNC machine tool operation data; use the isolated forest algorithm to obtain the outlier score of each data point in the sample space;

[0006] Based on the distribution characteristics of the data points in the sample space, the location factor of each data point is obtained;

[0007] Based on the anomaly score and location factor corresponding to each data point, and combined with a preset function, a correction weight is obtained; the corresponding anomaly score is corrected according to the correction weight to obtain the corrected anomaly score for each data point.

[0008] A neighborhood region for each data point is established using a preset method; based on the salience characteristics of the corrected anomaly score of the data point in the corresponding neighborhood region, and combined with the salience characteristics of the corrected anomaly score of the data point relative to the corrected anomaly scores of all data points, the anomaly probability of each data point is obtained.

[0009] Abnormal data points are filtered out based on the probability of anomaly for each data point.

[0010] Furthermore, the method for obtaining the position factor includes:

[0011] The sample space is a two-dimensional sample space; a two-dimensional coordinate system is established with each data point as the origin, and the first distribution parameter is obtained based on the difference in the number of data points in different quadrants;

[0012] Cluster the data points in the two-dimensional sample space; connect the cluster center of each cluster to other cluster centers, and obtain the second distribution parameter based on the variation characteristics of the angle formed by the lines connecting each cluster center to other cluster centers; the angle is the angle formed by adjacent connecting lines.

[0013] The Euclidean distance from each data point to the centroid of all samples in the two-dimensional sample space is normalized and used as the third distribution parameter for each data point.

[0014] The location factor of each data point is obtained based on the first distribution parameter, the second distribution parameter, and the third distribution parameter corresponding to each data point; the first distribution parameter, the second distribution parameter, and the third distribution parameter are all positively correlated with the location factor.

[0015] Furthermore, the method for obtaining the first distribution parameter includes:

[0016] The absolute value of the difference between the number of data points in two different quadrants is used as the quadrant difference parameter for the two corresponding quadrants; the sum of all the quadrant difference parameters corresponding to each data point is used as the first distribution parameter for each data point.

[0017] Furthermore, the method for obtaining the second distribution parameter includes:

[0018] The included angles with each cluster center as the vertex are sorted in descending order of angle value to obtain the included angle sequence corresponding to each cluster center; the mean of the ratios of adjacent included angles in the included angle sequence is negatively correlated and used as the second distribution parameter; the included angle with the smallest sequence number among the ratios of adjacent included angles is used as the denominator.

[0019] Furthermore, the method for obtaining the position factor includes:

[0020] The product of the first distribution parameter, the second distribution parameter, and the third distribution parameter corresponding to each data point is normalized and used as the location factor for each data point.

[0021] Furthermore, the method for obtaining the corrected weights includes:

[0022] The preset function is the probability density function of the standard normal distribution; after mapping the product of the outlier score and the location factor corresponding to each data point to the preset symmetric value range, the mapped value is substituted into the probability density function of the standard normal distribution, and the function value is normalized and then a constant 1 is added to obtain the corrected weight; the center of symmetry of the preset symmetric value range is the origin.

[0023] Furthermore, the method for obtaining the corrected abnormal score includes:

[0024] The product of the anomaly score and the correction weight for each data point is used as the corrected anomaly score for each data point.

[0025] Furthermore, the method for obtaining the probability of an anomaly includes:

[0026] The Euclidean distance between each data point and its nearest other data points is obtained as a reference distance, and the average of the reference distances corresponding to all data points is used as the neighborhood radius parameter; the product of the preset radius parameter and the neighborhood radius parameter is used as the neighborhood radius, and a neighborhood region is established with each data point as the center; the preset radius parameter is a positive number.

[0027] The absolute value of the difference between the corrected outlier score of each data point and the mean of the corrected outlier scores of all data points in its neighborhood is used as the local salience parameter for each data point.

[0028] The ratio of the corrected outlier score for each data point to the maximum corrected outlier score across all data points is used as the global salience parameter for each data point.

[0029] Based on the local salience parameter and the global salience parameter of each data point, the anomaly probability of each data point is obtained; the local salience parameter is negatively correlated with the anomaly probability; the global salience parameter is positively correlated with the anomaly probability.

[0030] Furthermore, the method for obtaining the probability of an anomaly includes:

[0031] The ratio of the global salience parameter to the local salience parameter for each data point is normalized and used as the anomaly probability for each data point.

[0032] Furthermore, the method for obtaining the abnormal data points includes:

[0033] Data points whose probability of being abnormal is greater than or equal to a preset abnormal threshold are marked as abnormal data points.

[0034] The present invention has the following beneficial effects:

[0035] This invention first acquires CNC machine tool operation data to provide a data foundation for subsequent analysis; establishes a sample space to prepare for analyzing the distribution characteristics of data points; uses the isolated forest algorithm to obtain anomaly scores for data points, preparing for subsequent correction of these scores; further, based on the distribution characteristics of data points in the sample space, obtains the location factor for each data point, providing distribution characteristic parameters to facilitate subsequent correction of anomaly scores based on these distribution characteristics; further, based on the anomaly score and location factor corresponding to each data point, and combined with a preset function, obtains correction weights to facilitate the stretching and amplification of anomaly scores for locally clustered anomalies, better reflecting their own anomaly characteristics, thus enabling more accurate anomaly evaluation in subsequent steps; further, corrects the corresponding anomaly scores based on the correction weights, analyzing the local and global salience characteristics of the corrected anomaly scores for each data point to obtain a more accurate anomaly probability for each data point, preparing for the screening of anomaly data points; finally, based on the anomaly probability of each data point, screens out anomaly data points, facilitating analysis by relevant personnel. This invention addresses the problem that traditional isolated forest algorithms are inaccurate in evaluating outlier data with local clustering characteristics. By analyzing the distribution characteristics of data points and combining local and global salience features to correct outlier scores, it can more accurately evaluate data points, filter out more accurate outlier data points, improve production efficiency, and ensure product quality. Attached Figure Description

[0036] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 A flowchart illustrating an intelligent analysis method for CNC machine tool operation data provided in one embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram of a sample space for constructing CNC machine tool operation data, provided as an embodiment of the present invention. Detailed Implementation

[0039] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a CNC machine tool operation data intelligent analysis method proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0040] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0041] The following description, in conjunction with the accompanying drawings, details the specific scheme of the intelligent analysis method for CNC machine tool operation data provided by the present invention.

[0042] This invention is based on the Isolation Forest algorithm. First, the Isolation Forest algorithm is used to obtain the outlier scores of data points. Then, the distribution characteristics of data points in the sample space are analyzed, and the outlier scores of data points that are distributed at the edge of the sample space but have local clustering characteristics are corrected. Furthermore, the salience characteristics of the corrected outlier scores of data points in local and global contexts are combined to obtain the probability of data points being outliers, thereby obtaining outlier data points.

[0043] Please see Figure 1 The diagram illustrates a flowchart of an intelligent analysis method for CNC machine tool operation data according to an embodiment of the present invention, specifically including:

[0044] Step S1: Obtain CNC machine tool operation data; establish a sample space based on the CNC machine tool operation data; use the isolated forest algorithm to obtain the outlier score of each data point in the sample space.

[0045] In this embodiment of the invention, CNC machine tool operation data is first acquired to provide a data foundation for subsequent analysis; a sample space is established to prepare for subsequent analysis of the distribution characteristics of data points; at the same time, the isolated forest algorithm is used to obtain the anomaly scores of data points to prepare for subsequent correction of the anomaly scores of data points and to obtain a more accurate anomaly probability.

[0046] It should be noted that, in one embodiment of the present invention, by setting the sampling interval to 10 seconds, temperature data and vibration frequency data of a certain component of a CNC machine tool are collected within one operating cycle to construct a two-dimensional sample space. Please refer to [link to relevant documentation]. Figure 2This illustration shows a schematic diagram of a sample space constructed from CNC machine tool operation data according to an embodiment of the present invention. The horizontal axis represents temperature, and the vertical axis represents vibration frequency. Each data point in the sample space is a coordinate point determined by the temperature data and vibration frequency data corresponding to a sampling time. In other embodiments of the present invention, the quantity, type, and sampling frequency of the operation data can be adjusted according to actual needs. Implementers can construct a multi-dimensional sample space from multiple types of data, but it is also necessary to consider the impact on algorithm performance and the increase in computational complexity as the dimensions of the data increase. Alternatively, existing technologies such as the DTW algorithm can be used to analyze the correlation between different data, placing the data with the strongest correlation in the same sample space for analysis, or pairing them up and placing them in the same sample space for analysis.

[0047] It should be noted that the isolated forest algorithm is a well-known technique in the field, and will not be elaborated upon here.

[0048] Step S2: Obtain the location factor of each data point based on the distribution characteristics of the data points in the sample space.

[0049] In this embodiment of the invention, considering that during normal operation of a CNC machine tool, the data points in the sample space are concentrated in the central area, while abnormal data, due to component failures or environmental factors, are distributed at the edges of the sample space and are generally relatively discrete. However, if a certain failure persists for a long time, then locally dense abnormal data may occur at the edges of the sample space, potentially resulting in... Figure 2 Regions A and B, as shown, are located at the edge of the sample space and differ from the large amount of data clustered in the central region. However, regions A and B themselves have a certain degree of clustering. The Isolation Forest algorithm mainly considers the degree of isolation of data points. When abnormal data clusters occur, the abnormal scores of these abnormal data will be affected, resulting in inaccurate analysis of CNC machine tool operation data. Therefore, it is also necessary to obtain the position factor of each data point based on the distribution characteristics of the data points in the sample space and correct the abnormal scores.

[0050] It should be noted that a large amount of data in the sample space may not be concentrated in the central region, but may be in a fixed region, such as the region enclosed by the normal temperature range and the normal vibration frequency range in one embodiment of the present invention. However, through certain transformations, such as overall translation, the region enclosed by the normal temperature range and the normal vibration frequency range can be moved to the center of the sample space. Then the normal data will be concentrated in the central region, while not changing the relative position of the data points in the sample space and not affecting the distribution characteristics of the data points.

[0051] It should be noted that the central or middle position is where a large number of data points are clustered, such as... Figure 2The region C shown represents the area corresponding to the normal temperature range and the normal vibration frequency range.

[0052] Preferably, in one embodiment of the present invention, considering that the closer the data point is to the edge of the sample space, the greater the difference between data points in different quadrants when a two-dimensional coordinate system is established with each data point as the origin, the absolute value of the difference between the number of data points in two different quadrants is used as the quadrant difference parameter of the corresponding two quadrants; and the sum of all quadrant difference parameters corresponding to each data point is used as the first distribution parameter of each data point.

[0053] Furthermore, considering the clustering of data points within the sample space, the closer a cluster is to the edge of the sample space, the closer the data points within that cluster are to the edge of the sample space. The closer a cluster center is to the center of the sample space, the more divergent the lines connecting it to other cluster centers, and the smaller the difference in the angles between adjacent lines. Conversely, the farther away the lines are, the stronger the clustering. Correspondingly, with the cluster center as the vertex, the angles between adjacent lines change; the larger the maximum angle for a cluster center closer to the edge, and the worse the uniformity of the angles. Therefore, we cluster the data points in the two-dimensional sample space. We connect the cluster centers of each cluster to other cluster centers, and sort the angles with each cluster center as the vertex in descending order of angle value to obtain the angle sequence corresponding to each cluster center. We then perform a negative correlation mapping on the mean of the ratios of adjacent angles in the angle sequence, using this as the second distribution parameter. The angle with the smallest index among the ratios of adjacent angles is used as the denominator.

[0054] Meanwhile, considering that the greater the distance from a data point to the centroid of all data points in the sample space, the more likely the data point is to be located in the edge region of the sample space, the Euclidean distance from each data point to the centroid of all samples in the two-dimensional sample space is normalized and used as the third distribution parameter for each data point.

[0055] The product of the first, second, and third distribution parameters corresponding to each data point is normalized and used as the location factor for each data point.

[0056] The formulas for calculating the location factor include:

[0057] ;

[0058] in, Indicates the sequence number of the data point; Indicates the first Location factor of each data point; This represents the standard normalization function; Indicates the first The first distribution parameter of each data point , The index indicates the number of pairs of combinations within each quadrant. The number of types that can be paired and combined within a quadrant is indicated in one embodiment of the present invention. ; Indicates the first The second distribution parameter of each data point , This indicates the index of the included angle in the included angle sequence. Indicates the first The number of included angles in the included angle sequence corresponding to each data point's cluster. Indicates the first The corresponding data point is the th The angle value of the included angle, Indicates the first The corresponding data point is the th The angle value of each included angle; Indicates the first The third distribution parameter of each data point , Indicates the first The Euclidean distance from each data point to the centroid of all other data points. This represents the normalization function for maximum and minimum values.

[0059] In the formula for calculating the location factor, a larger first distribution parameter indicates a more uneven distribution of data points across different quadrants of the coordinate system centered on the data point, more closely resembling the characteristics of data points distributed at the edge of the sample space, thus resulting in a larger location factor; a larger second distribution parameter indicates... The smaller, The smaller the value, the worse the uniformity of the angle, indicating that the cluster center is more deviated from other cluster centers, and the more likely the data points within the cluster are to be at the edge of the sample space, resulting in a larger location factor. The larger the third distribution parameter, the farther the data point is from the centroid of all data points in the sample space, indicating that the data point is more likely to be in the edge region of the sample space, resulting in a larger location factor. Obtaining the location factor from three perspectives using the first, second, and third distribution parameters improves the accuracy of the location factor and is more conducive to the subsequent correction of outlier scores.

[0060] It should be noted that in other embodiments of the present invention, other basic mathematical operations or function mappings may also be used to implement the relevant mappings. These are all technical means well known to those skilled in the art, and will not be described in detail here.

[0061] It should be noted that the first distribution parameter, the second distribution parameter, and the third distribution parameter can all represent the distribution characteristics of data points in the sample space. In other embodiments of the present invention, the location factor can be obtained by retaining only one or any two of the three parameters, thereby improving the analysis efficiency by sacrificing some accuracy.

[0062] Step S3: Based on the outlier score and location factor corresponding to each data point, and combined with a preset function, obtain the correction weight; correct the corresponding outlier score according to the correction weight to obtain the corrected outlier score for each data point.

[0063] In this embodiment of the invention, considering that ordinary anomalous data is distributed at the edge of the sample space and is relatively discrete, the anomalous score and location factor are both large, and the overall value after weighting by directly using the location factor will be larger; while anomalous data with local clustering characteristics has a more similar anomalous score to normal data, but the location factor of anomalous data with local clustering characteristics is larger than that of normal data; at the same time, the anomalous score and location factor of normal data are both small, and this embodiment of the invention aims to stretch and amplify the anomalous score of locally clustered anomalous data through weighting to reflect its own anomalous characteristics. Therefore, it is necessary to obtain a correction weight based on the anomalous score and location factor corresponding to each data point, combined with a preset function, in order to prepare for obtaining a more accurate corrected anomalous score.

[0064] Preferably, in one embodiment of the present invention, considering that the anomaly score and location factor of ordinary abnormal data are both large, the product of the anomaly score and location factor is large; the anomaly score and location factor of normal data are both small, the product of the anomaly score and location factor is small; and the anomaly score of abnormal data with local clustering characteristics is small, while the location factor is large, so the product of the anomaly score and location factor is neither as large as the product corresponding to ordinary abnormal data nor as small as the product corresponding to normal data, and is at a medium level. In order to minimize the impact on the anomaly scores of ordinary and normal data, while amplifying the anomaly scores of abnormal data with local clustering characteristics, and since the probability density function of the standard normal distribution can just match the corresponding trend, the probability density function of the standard normal distribution is used as the preset function; after mapping the product of the anomaly score and location factor corresponding to each data point to the preset symmetric value range, the mapped value is substituted into the probability density function of the standard normal distribution, and the function value is normalized and a constant 1 is added to obtain the corrected weight.

[0065] The formulas for calculating the adjusted weights include:

[0066] ;

[0067] in, Indicates the sequence number of the data point; Indicates the first The adjusted weights for each data point; 1 is a constant; This represents the standard normalization function; Pi; It is a natural constant; Indicates the first Location factor of each data point; Indicates the first The outlier score of each data point; This represents a preset mapping function used to map the product of the outlier score and the position factor to a preset symmetric range.

[0068] It should be noted that, in one embodiment of the present invention, the preset symmetric range is [-1, 1], and the function used to map the product of the anomaly score and the position factor to [-1, 1] is: In other embodiments of the present invention, the implementer may also use other preset functions for mapping. For example, the probability density function of the Student's t-distribution of the standard normal distribution, which is similar to the probability density function of the standard normal distribution, may be selected as the preset function. The parameters in the preset function may also be adjusted to make the curve of the probability density function flatter or steeper. Other preset symmetric ranges may also be set. When mapping the product, other preset mapping functions may also be used to better adapt to the needs of the implementation scenario. These are all technical means well known to those skilled in the art and will not be described in detail here.

[0069] After obtaining the corrected weights for the outlier scores of the data points, the outlier scores can be corrected.

[0070] Preferably, in one embodiment of the present invention, the product of the anomaly score of each data point and the correction weight is used as the corrected anomaly score of each data point.

[0071] Step S4: Establish a neighborhood region for each data point using a preset method; based on the salience characteristics of the corrected anomaly score of the data point in the corresponding neighborhood region, and combined with the salience characteristics of the corrected anomaly score of the data point relative to the corrected anomaly scores of all data points, obtain the anomaly probability of each data point.

[0072] In this embodiment of the invention, considering that the larger the corrected anomaly score, the more likely the corresponding anomaly score is to be abnormal production data, and the greater the salience among the anomaly scores of all data points globally; the more similar the anomaly score of a data point is to the anomaly scores of data points in its neighborhood, the weaker the local salience feature, indicating that the anomaly score of the data point is more reliable and less likely to be noise, in order to more accurately filter out possible abnormal data, the salience feature of the corrected anomaly score of the data point in its corresponding neighborhood, combined with the salience feature of the corrected anomaly score of the data point relative to the corrected anomaly scores of all data points, is used to obtain the anomaly probability of each data point.

[0073] Preferably, in one embodiment of the present invention, considering that the greater the difference between the corrected anomaly score of a data point and the mean of the corrected anomaly scores of data points in its neighborhood, the greater the local salience of the corrected anomaly score of the data point, the more likely it is a noise point, and the less likely it is an anomalous data point, and the less likely it is to be an anomalous; considering that the greater the ratio of the corrected anomaly score of a data point to the maximum corrected anomaly score, the greater the global salience of the corrected anomaly score of the data point, the more likely it is to be an anomalous data point, and the more likely it is to be an anomalous.

[0074] Based on this, the Euclidean distance between each data point and its nearest other data points is obtained as the reference distance, and the average of the reference distances corresponding to all data points is used as the neighborhood radius parameter; the product of the preset radius parameter and the neighborhood radius parameter is used as the neighborhood radius, and a neighborhood region is established with each data point as the center.

[0075] The absolute value of the difference between the corrected outlier score of each data point and the mean of the corrected outlier scores of all data points in its neighborhood is used as the local salience parameter for each data point.

[0076] The ratio of the corrected outlier score of each data point to the maximum corrected outlier score among all data points is used as the global salience parameter for each data point.

[0077] The ratio of the global salience parameter to the local salience parameter for each data point is normalized and used as the anomaly probability for each data point.

[0078] The formulas for calculating the probability of an anomaly include:

[0079] ;

[0080] in, Indicates the sequence number of the data point; Indicates the first The probability of anomalies in each data point; express function; Indicates the first Corrected outlier scores for each data point; Indicates the sequence number of the data point within the neighborhood area; Indicates the first The number of data points within the neighborhood of each data point; Indicates the first Within the neighborhood of the data point, the first... Corrected outlier scores for each data point; This represents the maximum corrected outlier score for all data points. This represents a function to obtain the maximum value. Indicates the first Global highlighting parameters for each data point; Indicates the first Local salience parameters for each data point.

[0081] It should be noted that in other embodiments of the present invention, other basic mathematical operations or function mappings may also be used to implement the relevant mappings. These are all technical means well known to those skilled in the art, and will not be described in detail here.

[0082] It should be noted that in one embodiment of the present invention, the preset radius parameter is 10. In other embodiments of the present invention, the implementer can set the preset radius parameter to adjust the neighborhood radius, or the neighborhood region can be established by using a preset neighborhood radius; considering the calculation formula for the probability of anomalies... In some cases, the value may be zero. This indicates that the corresponding data point is relatively isolated in the sample space and there are no other data points in the neighborhood. In order to avoid missing abnormal data points, the probability of its abnormality is set to a maximum of 1.

[0083] Step S5: Filter out abnormal data points based on the probability of anomalies for each data point.

[0084] In this embodiment of the invention, after obtaining the probability of anomalies that can accurately characterize the degree of anomalies of data points, abnormal data points can be screened.

[0085] Preferably, in one embodiment of the present invention, considering that the greater the probability of an anomaly of a data point, the more likely it is to be an abnormal data point, data points with an anomaly probability greater than or equal to a preset anomaly threshold are marked as abnormal data points.

[0086] In one embodiment of the present invention, the preset abnormal threshold is 0.88. In other embodiments of the present invention, the implementer may adjust it according to the actual implementation scenario.

[0087] Once abnormal data points are obtained, relevant personnel can analyze the timing and characteristics of these abnormal data points, analyze the working status of the CNC machine tool, and adjust working parameters or perform maintenance in a timely manner to improve production efficiency and ensure product quality.

[0088] In summary, to address the technical problem of inaccurate analysis of CNC machine tool operation data using traditional isolated forest algorithms, this invention provides an intelligent analysis method for CNC machine tool operation data. This invention first acquires CNC machine tool operation data, establishes a sample space, and uses the isolated forest algorithm to obtain anomaly scores for data points. Further, based on the distribution characteristics of data points in the sample space, it obtains the location factor for each data point. Then, based on the anomaly score and location factor corresponding to each data point, and combined with a preset function, it obtains a correction weight. The anomaly score is corrected according to the correction weight to obtain a corrected anomaly score for each data point. Next, it analyzes the salience characteristics of data points to obtain the anomaly probability of each data point. Finally, based on the anomaly probability of each data point, it filters out anomaly data points. This invention addresses the problem of inaccurate evaluation of anomaly data with local clustering characteristics using traditional isolated forest algorithms. By analyzing the distribution characteristics of data points and combining the local and global salience characteristics of the corrected anomaly scores, it provides a more accurate anomaly evaluation of data points, filtering out more accurate anomaly data points to improve production efficiency and ensure product quality.

[0089] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0090] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for intelligent analysis of operation data of a numerical control machine tool, characterized in that, The method comprises: acquiring numerical control machine tool operation data; establishing a sample space according to the numerical control machine tool operation data; and acquiring an anomaly score of each data point in the sample space by using an isolation forest algorithm; acquiring a position factor of each data point according to a distribution feature of the data point in the sample space; obtaining a correction weight according to the anomaly score and the position factor corresponding to each data point and in combination with a preset function; correcting the anomaly score corresponding to each data point according to the correction weight to obtain a corrected anomaly score of each data point; establishing a neighborhood region of each data point in a preset manner; obtaining an anomaly possibility of each data point according to a highlighting feature of the corrected anomaly score of the data point in the corresponding neighborhood region and in combination with a highlighting feature of the corrected anomaly score of the data point relative to the corrected anomaly scores of all data points; screening out an anomaly data point according to the anomaly possibility of each data point; The position factor is obtained by the following method: the sample space is a two-dimensional sample space; a two-dimensional coordinate system is established with each data point as an origin; a first distribution parameter is obtained according to a difference feature of the number of data points in different quadrants; the data points in the two-dimensional sample space are clustered; the cluster centers of each cluster are connected with other cluster centers; a second distribution parameter is obtained according to a variation feature of an included angle formed by the connecting lines of each cluster center and other cluster centers; the included angle is an included angle formed by adjacent connecting lines; a third distribution parameter of each data point is obtained by normalizing the Euclidean distance of each data point to the centroid of all samples in the two-dimensional sample space; the position factor of each data point is obtained according to the first distribution parameter, the second distribution parameter and the third distribution parameter corresponding to each data point; the first distribution parameter, the second distribution parameter and the third distribution parameter are all positively correlated with the position factor; The correction weight is obtained by the following method: the preset function is a probability density function of a standard normal distribution; a product of the anomaly score and the position factor corresponding to each data point is mapped to a preset symmetric value range; the mapped value is brought into the probability density function of the standard normal distribution; the function value is normalized and then a constant 1 is added to obtain the correction weight; and a symmetric center of the preset symmetric value range is the origin.

2. The intelligent analysis method for operation data of a CNC machine tool according to claim 1, characterized in that, The first distribution parameter is obtained by the following method: The absolute value of the difference between the number of data points in two different quadrants is taken as a quadrant difference parameter of the two quadrants; and the sum of all the quadrant difference parameters corresponding to each data point is taken as the first distribution parameter of each data point.

3. The intelligent analysis method of the operation data of the numerical control machine tool according to claim 1, characterized in that, The second distribution parameter is obtained by the following method: The included angles with each cluster center as a vertex are sorted in descending order according to the angle values to obtain an included angle sequence corresponding to each cluster center; a mean value of the ratio of adjacent included angles in the included angle sequence is negatively correlated to obtain a second distribution parameter; and the included angle with the smallest serial number in the ratio of adjacent included angles is taken as a denominator.

4. The intelligent analysis method of the operation data of the numerical control machine tool according to claim 1, characterized in that, The position factor is obtained by the following method: The product of the first distribution parameter, the second distribution parameter and the third distribution parameter corresponding to each data point is normalized as a position factor of each data point.

5. The intelligent analysis method of the numerical control machine tool operation data according to claim 1, characterized in that, The method for obtaining the modified abnormal score comprises: The product of the abnormal score of each data point and the modified weight is taken as a modified abnormal score of each data point.

6. The intelligent analysis method of the numerical control machine tool operation data according to claim 1, characterized in that, The method for obtaining the abnormal possibility comprises: The Euclidean distance between each data point and its nearest other data point is taken as a reference distance, and the average of the reference distances corresponding to all data points is taken as a neighborhood radius parameter; the product of a preset radius parameter and the neighborhood radius parameter is taken as a neighborhood radius, and a neighborhood region is established with each data point as the center; the preset radius parameter is a positive number; The absolute value of the difference between the modified abnormal score of each data point and the average of the modified abnormal scores of all data points in the neighborhood region of each data point is taken as a local highlight parameter of each data point; The ratio of the modified abnormal score of each data point to the maximum of the modified abnormal scores of all data points is taken as a global highlight parameter of each data point; The abnormal possibility of each data point is obtained according to the local highlight parameter and the global highlight parameter of each data point; the local highlight parameter is negatively correlated with the abnormal possibility; and the global highlight parameter is positively correlated with the abnormal possibility.

7. The intelligent analysis method of the operation data of a CNC machine tool according to claim 6, characterized in that, The method for obtaining the abnormal possibility comprises: The ratio of the global highlight parameter and the local highlight parameter of each data point is normalized as the abnormal possibility of each data point.

8. The intelligent analysis method of the operation data of a CNC machine tool according to claim 1, characterized in that, The method for obtaining the abnormal data point comprises: Data points with an abnormal possibility greater than or equal to a preset abnormal threshold are marked as abnormal data points.

Images