A milling machine processing intelligent monitoring method and system

Abstract

The application belongs to the technical field of intelligent monitoring, and particularly relates to a milling machine processing intelligent monitoring method and system, which comprises the following steps: collecting data values of each monitoring dimension at all historical moments in the current milling machine processing process, constructing an original data set and clustering; determining a suspected abnormal working condition cluster based on the density and dynamic trend of each clustering cluster, and reconstructing the data set to calculate the reconstructed abnormal score of each data point, while calculating the original abnormal score of each data point in the original data set, taking the maximum value of the two to generate a warning coefficient, and using the same to evaluate the processing state to realize intelligent monitoring. The application solves the problem that traditional algorithms are difficult to identify abnormal data dense distribution by constructing a double-channel monitoring model and intelligently reconstructing the data set, and significantly improves the accuracy of monitoring and warning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent monitoring. More particularly, the present application relates to a milling machine processing intelligent monitoring method and system. BACKGROUND

[0002] The stability of the milling process is crucial to ensure product quality and equipment life. Typical abnormal working conditions, such as inappropriate equipment parameter settings, cause tool wear to intensify or the machine tool to vibrate violently. If not addressed in a timely manner, it will lead to a decline in workpiece surface quality, and even cause tool breakage and other unsafe incidents. Therefore, monitoring abnormal conditions during the milling process is an urgent need in the field of intelligent manufacturing.

[0003] Traditional methods for monitoring abnormal conditions during the milling process, such as the Isolation Forest algorithm, are used to detect abnormal data points in the milling process. The basic principle is that abnormal points are easily isolated due to their rarity and difference.

[0004] However, this principle fails when faced with specific abnormalities in the milling process, resulting in a decline in detection accuracy. For example, when the process enters the initial vibration state, the vibration signal will show a steady and slow increase. At this time, the feature data points collected represent extreme abnormal working conditions, but they form a relatively dense cluster in the feature space because of their similar differences. For traditional Isolation Forest algorithms, these relatively dense data points are difficult to isolate quickly, and their path length will be longer, resulting in a lower abnormal score. Ultimately, they will be misjudged as normal conditions. This defect of identifying persistent abnormalities as normal leads to the system being unable to provide timely and effective early warnings for high-risk conditions such as violent vibration and tool wear. SUMMARY

[0005] To solve the technical problem of the failure of the monitoring algorithm due to the dense distribution of abnormal data, the present application provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides an intelligent monitoring method for milling machine machining, comprising: collecting data values ​​of each dimension at all historical moments during the current milling process, obtaining data points at each historical moment, and constructing an original dataset; clustering all data points in the original dataset to obtain several clusters; determining the density of each cluster based on the distribution of distances between data points within each cluster; selecting clusters representing stable cutting states based on the density; determining the necessity of each cluster as part of the overall dataset based on the trend changes between each data point within each cluster and the cluster centers of the stable cutting state clusters; and calculating a dynamic threshold based on the necessity. The algorithm calculates the reconstructed anomaly score for each data point in a cluster and divides the clusters into suspected abnormal operating condition clusters and suspected normal operating condition clusters. It then constructs a new dataset by combining the cluster centers of all suspected abnormal operating condition clusters with the data points within all suspected normal operating condition clusters. The new dataset is then input into the Isolation Forest algorithm to determine the reconstructed anomaly score for each data point. The reconstructed anomaly score of the cluster center of the suspected abnormal operating condition cluster is assigned to all data points within that cluster. The original dataset is then input into the Isolation Forest algorithm to obtain the original anomaly score for each data point. The maximum value between the reconstructed anomaly score and the original anomaly score is used as the warning coefficient for each data point. Finally, control and warning systems are implemented based on the warning coefficients of all historical data points.

[0007] This invention performs cluster analysis on data from the current milling process, identifying suspected abnormal operating condition clusters based on the density and dynamic trends of the clusters. It also introduces dataset reconstruction technology to enhance the isolation of these abnormal clusters in the isolated forest algorithm. Finally, it combines the monitoring results of the reconstructed data and the original data for dual verification and early warning, effectively solving the technical problem of missed detection in the traditional isolated forest algorithm when dealing with early potential abnormal operating conditions with dense distribution characteristics. By simplifying dense clusters of abnormal points to their cluster centers to reconstruct the dataset, the sparsity of these anomalies in the feature space is greatly enhanced, enabling the algorithm to quickly identify them. This significantly improves the accuracy of monitoring the entire operating process and provides a solid and reliable guarantee for preventing workpiece quality degradation.

[0008] Preferably, the clustering of all data points in the original dataset is achieved through an iterative self-organizing clustering algorithm.

[0009] Preferably, the density of each cluster satisfies the expression: In the formula, For the first The density of each cluster; For the first Within the cluster, the first The data point and the Euclidean distance between data points; For the first The index values ​​of data points within each cluster, and ; For the first The total number of data points within each cluster; It is a natural exponential function; This is the index value of the cluster.

[0010] This invention evaluates the density of each cluster by calculating the average Euclidean distance between data points within the cluster, providing an objective mathematical basis for the subsequent automatic identification of the most stable steady cutting state cluster. This makes the benchmark determination process more automated and accurate without human intervention.

[0011] Preferably, the step of selecting stable cutting state clusters based on density includes: traversing the density of each cluster and selecting the cluster corresponding to the maximum density as the stable cutting state cluster.

[0012] Preferably, the necessity of each cluster as part of the overall dataset satisfies the expression: In the formula, For the first The necessity of individual clusters as part of the overall dataset; For the first Within the cluster, the first Cluster centers of data points and stable cutting state clusters The slope values ​​between; The cluster center of the stable cutting state cluster; For the first The index value and total number of data points within each cluster; This is the index value of the cluster; To determine the absolute value sign; This is the standard normalization function.

[0013] This invention provides a dynamic perspective on the trend of data state changes within a cluster by calculating the average slope change of data points within the cluster relative to the center of the cluster in a stable cutting state. This can effectively distinguish between stable normal clusters and continuously deteriorating abnormal clusters, thus improving the ability to identify early fault characteristics.

[0014] Preferably, the step of calculating a dynamic threshold based on necessity and dividing several clusters into suspected abnormal operating condition clusters and suspected normal operating condition clusters includes: traversing the necessity of each cluster as part of the overall dataset, the dynamic threshold being equal to the average of the maximum and minimum values ​​of the necessity of each cluster as part of the overall dataset; and marking clusters whose necessity is greater than the dynamic threshold as suspected abnormal operating condition clusters, and the remaining clusters as suspected normal operating condition clusters.

[0015] Preferably, the reconstruction anomaly score of each sample point in the new dataset satisfies the expression: In the formula, In the new dataset, the first Reconstruction anomaly score for each sample point; In the new dataset, the first The sample point and the first The total number of data points within the cluster of each sample point; The index value of the sample points in the new dataset; This represents the total number of sample points in the new dataset. In the new dataset, the maximum total number of data points within the cluster to which all sample points belong; In the new dataset, the first The sample point at the th th Path length in a single isolated tree; The index value and total number of the isolation tree; It is a natural exponential function.

[0016] In calculating the reconstructed anomaly score, this invention introduces the relative size of the original cluster as a weighting factor, enabling sample points of rarer operating condition clusters to obtain higher anomaly scores, thereby further enhancing the monitoring system's sensitivity to early but critical anomalies.

[0017] Preferably, the path length is obtained by inputting all sample points of the new dataset into an isolated forest algorithm to construct multiple isolated trees.

[0018] Preferably, the control and early warning based on the early warning coefficients of data points at all historical moments includes: traversing the early warning coefficients of data points at all historical moments during the current milling process; if the number of data points with early warning coefficients greater than a preset alarm threshold exceeds 10% of the total number of data points at all historical moments, it is determined that there is an abnormality in the current milling process, and control and early warning are issued.

[0019] This invention establishes an early warning triggering mechanism based on the proportion of abnormal points, which avoids false alarms caused by single or instantaneous spike data points, thereby enhancing the robustness of the entire monitoring system and reducing unnecessary early warnings that trigger incorrect parameter tuning.

[0020] Secondly, the present invention provides an intelligent monitoring system for milling machine machining, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned intelligent monitoring method for milling machine machining is implemented.

[0021] By adopting the above technical solution, a computer program is generated from the above-mentioned intelligent monitoring method for milling machine processing and stored in the memory so that it can be loaded and executed by the processor. In this way, a terminal device can be made based on the memory and the processor for convenient use.

[0022] The beneficial effects of this invention are as follows:

[0023] (1) The present invention introduces the dynamic evolution trend of data clusters as an evaluation dimension, which can quantitatively identify the state representing slow performance degradation, realize the evaluation of the progressive state of faults, and provide data support for identifying suspected abnormal operating condition clusters;

[0024] (2) By reconstructing the monitoring data, the present invention transforms the densely distributed progressive anomalies into sparse isolated features, which solves the problem of missed detection caused by the local similarity of abnormal data in traditional algorithms and significantly improves the ability to identify early faults.

[0025] (3) The present invention integrates the abnormal score monitoring results of reconstructed data and original data for early warning, which significantly reduces the false alarm rate and enhances the reliability of abnormal working condition early warning. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating an intelligent monitoring method for milling machine processing according to the present invention;

[0027] Figure 2 This is a schematic flowchart illustrating step S3 in a milling machine intelligent monitoring method according to the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0030] This invention discloses an intelligent monitoring method for milling machine machining, referring to... Figure 1 This includes steps S1-S5:

[0031] S1. Collect data values ​​for each dimension of all historical moments during the current milling process.

[0032] It should be noted that during milling machine operation, the operating status is manifested through physical signals such as vibration and spindle current. For example, vibration is stable during normal cutting, but as the tool gradually wears down, the vibration energy increases slowly, eventually leading to severe chatter. Under these conditions, the tool edge becomes dull, no longer efficiently cutting metal, but instead experiencing more compression, resulting in increased overall cutting resistance. To maintain the set speed, the spindle motor must output greater torque, and its operating current will continuously increase. Therefore, it is necessary to collect high-frequency data on various status parameters of the milling machine during machining to ensure that potential abnormal conditions are identified in a timely manner.

[0033] Specifically, an acceleration sensor is installed near the tool and workpiece fixture to collect vibration data during the machining process; a current transformer is installed in the machine tool spindle motor area to collect current data during the machining process; for example, the acquisition frequency of each sensor can be set to 5Hz, which can be adjusted by the implementer according to the actual situation.

[0034] Considering that the unit dimension and quantity range may affect the analysis results when analyzing multidimensional data, the data values ​​of each dimension at the time of real-time acquisition need to be normalized by maximum and minimum before storage to eliminate the influence of unit dimension.

[0035] At this point, the data values ​​for each dimension at each moment during the current milling process have been obtained.

[0036] S2. Construct data points for each historical moment using the data values ​​of each monitoring dimension at each historical moment, and construct the original dataset for clustering to obtain several clusters.

[0037] It should be noted that, considering the data points from different machining states during milling, they will naturally form different clusters in the multidimensional feature space. For example, data points under a stable cutting state will cluster together to form a normal cluster; while data points under wear or severe chattering states will cluster in different regions to form other clusters because they deviate significantly from the data points in the normal cluster. Therefore, clustering can be used to identify these data clusters, thereby providing a basis for identifying potential abnormal clusters in the future.

[0038] Specifically, data points for each moment are constructed using the data values ​​of each monitoring dimension at each moment; and all historical data points during the current milling process are used as the original dataset; in this embodiment of the invention, an iterative self-organizing clustering algorithm is used to cluster all data points in the original dataset during the current milling process, resulting in several clusters and the cluster center of each cluster; wherein, the iterative self-organizing clustering algorithm optimizes the clustering results through an iterative process, so that the cluster centers can better represent the features in the dataset. The iterative self-organizing clustering algorithm is a well-known technology and will not be described in detail here.

[0039] Thus, we have obtained several clusters and the cluster center of each cluster during the current milling process.

[0040] S3. Determine the necessity of each cluster as part of the overall dataset.

[0041] Refer to the flowchart for step S3 Figure 2 The process includes steps S301 to S302, specifically as follows:

[0042] S301. Determine the density of each cluster based on the distribution of distances between data points within each cluster; based on the density, select clusters in a stable cutting state.

[0043] It should be noted that although the obtained non-stationary cutting state clusters represent abnormal working conditions, their internal data points are relatively dense due to their stable and continuous characteristics, which may affect the accuracy of subsequent potential anomaly identification. Therefore, further analysis of the clustering results is needed to identify clusters that may belong to non-stationary cutting states and treat them as a whole dataset to improve the accuracy of subsequent anomaly identification. Considering that the machining parameters are relatively stable under actual stable cutting conditions, the density of the clusters can be measured by analyzing the changes in the distance between all data points within the clusters, thereby obtaining the stable cutting state clusters.

[0044] Specifically, the density of each cluster is determined based on the distribution of distances between data points within each cluster; the density satisfies the expression:

[0045]

[0046] In the formula, For the first The density of each cluster; For the first Within the cluster, the first The data point and the Euclidean distance between data points; For the first The index values ​​of data points within each cluster, and ; For the first The total number of data points within each cluster; It is a natural exponential function; This is the index value of the cluster.

[0047] in, This value reflects the density of data points within the current cluster. The smaller the value, the closer the spatial distance between data points within the current cluster, indicating that the machining state parameters within the current cluster are more stable, and thus more likely to represent a cluster with a stable cutting state. Conversely, a larger value indicates that it is more likely to represent a cluster with a non-stationary cutting state. It should be noted that the density calculation formula applies to clusters with a total number of data points greater than or equal to 2. For clusters with a total number of data points less than 2, the density is defined as a preset value of 0 to ensure that they do not participate in the subsequent screening of stable cutting state clusters.

[0048] Thus, the density of each cluster during the current milling process can be obtained. Considering the stable characteristics of machining parameters under actual smooth cutting conditions, that is, the data points in the clusters corresponding to the smooth cutting conditions will be more densely distributed, the density of each cluster during the current milling process is traversed, and the cluster corresponding to the maximum density is taken as the smooth cutting state cluster.

[0049] S302. Based on the trend changes between each data point within each cluster and the cluster center of the stable cutting state cluster, determine the necessity of each cluster as a whole dataset.

[0050] It should be further explained that, considering the abnormal clusters of working conditions under non-stationary cutting conditions, the data points within them are relatively dense in the anomaly identification algorithm due to their steady and continuous growth characteristics, but they are indeed abnormal working conditions. Therefore, it is necessary to analyze the trend changes of the data points within each cluster and the index data points of the clusters under stationary cutting conditions to determine the necessity of using them as the overall dataset.

[0051] Specifically, based on the trend changes between each data point within each cluster and the cluster centers of the stable cutting state clusters, the necessity of each cluster as part of the overall dataset is determined; the necessity satisfies the expression:

[0052]

[0053] In the formula, For the first The necessity of individual clusters as part of the overall dataset; For the first Within the cluster, the first Cluster centers of data points and stable cutting state clusters The slope values ​​between; The cluster center of the stable cutting state cluster; For the first The index value and total number of data points within each cluster; This is the index value of the cluster; To determine the absolute value sign; This is the standard normalization function.

[0054] in, This value reflects the average trend of all data points in the current cluster relative to the cluster center of the stable cutting state cluster. The larger the value, the more likely all data points in the current cluster are on a trajectory of continuous movement away from the cluster center of the stable cutting state cluster, and the higher the necessity of it as a whole dataset. Conversely, the closer the value is to 0, the more likely all data points in the current cluster are in a stable state around the cluster center of the stable cutting state cluster, rather than a continuously deteriorating process, and the lower the necessity of it as a whole dataset.

[0055] Thus, we have determined the necessity of each cluster as part of the overall dataset during the current milling process.

[0056] S4. Based on the necessity calculation of the dynamic threshold, divide several clusters into suspected abnormal operating condition clusters and suspected normal operating condition clusters; construct a new dataset by combining the cluster centers of all suspected abnormal operating condition clusters with the data points in all suspected normal operating condition clusters; input the new dataset into the Isolation Forest algorithm to determine the reconstruction anomaly score of each sample point in the new dataset.

[0057] It should be noted that, considering the clusters in the initial state of abnormal operating conditions, although each data point exhibits a slow increase, they form a relatively dense cluster in the feature space due to their similar changes. To address the problem that traditional isolated forest algorithms struggle to identify dense abnormal data points, a dynamic threshold can be set to determine the necessity of each cluster as part of the overall dataset at the current moment. This distinguishes clusters that may belong to abnormal operating conditions, extracts their cluster centers as sample points for isolated forest anomaly detection, and makes the abnormal operating condition sample points very sparse in the new dataset. They are easily separated in each random cut, resulting in a shorter average path length in the isolation tree. In contrast, a large number of normal data points are densely distributed together, requiring many cuts to separate them one by one, resulting in a long average path length. Through this dataset reconstruction process, the accuracy of potential anomaly identification is significantly improved.

[0058] Specifically, the dynamic threshold in the current milling process is calculated, and the dynamic threshold is equal to the average of the maximum and minimum values ​​of the necessity of each cluster as part of the overall dataset. Preferably, the dynamic threshold can be obtained by the maximum inter-class variance method, specifically as follows: the necessity of each cluster as part of the overall dataset is used as a candidate threshold; based on the candidate threshold, the necessity of all clusters is divided into two groups; the inter-class variance between these two groups is calculated; the candidate threshold that maximizes the inter-class variance is found and used as the dynamic threshold in the current milling process; the necessity of each cluster as part of the overall dataset is traversed; clusters whose necessity is greater than the dynamic threshold in the current milling process are marked as clusters suspected of abnormal working conditions, and the remaining clusters are marked as clusters suspected of normal working conditions.

[0059] A new dataset is constructed by combining the cluster centers of all suspected abnormal operating condition clusters and the data points within all suspected normal operating condition clusters during the current milling process. All sample points in the new dataset are then input into the Isolation Forest algorithm to construct multiple isolation trees, and the path length of each sample point within each isolation tree is obtained. In this embodiment of the invention, the number of isolation trees can be set to 10, or adjusted according to actual conditions. The construction of isolation trees is a well-known technique and will not be elaborated upon here.

[0060] Furthermore, based on the path length of each sample point in each isolation tree in the new dataset and the total number of data points within the cluster to which each sample point belongs, a reconstruction anomaly score is determined for each sample point; the anomaly score satisfies the expression:

[0061]

[0062] In the formula, In the new dataset, the first Reconstruction anomaly score for each sample point; In the new dataset, the first The sample point and the first The total number of data points within the cluster of each sample point; The index value of the sample points in the new dataset; This represents the total number of sample points in the new dataset. In the new dataset, the maximum total number of data points within the cluster to which all sample points belong; In the new dataset, the first The sample point at the th th Path length in a single isolated tree; The index value and total number of the isolation tree; It is a natural exponential function.

[0063] in, Reflecting the new dataset, the first The isolation of a sample point is indicated by this value. The smaller the value, the easier it is for the current sample point to be separated and the closer it is to the root node. This means that the current sample point is more isolated in the constructed dataset. Reflecting the The relative size of the original cluster to which each sample point belongs; the smaller this value, the fewer data points there are within the original cluster to which the current sample point belongs, meaning that the processing state corresponding to the current sample point appears very infrequently in the entire historical data; therefore, through... right Make adjustments. The closer the value is to 1, the more normal the processing state of the current sample point is considered to be, and the original state should be maintained. That's all; and The closer the value is to 0, the more abnormal the processing state is considered to be at the current sample point, and the value should be significantly reduced accordingly. The above adjustments significantly improved the reconstruction anomaly score under abnormal operating conditions.

[0064] At this point, the reconstruction anomaly scores for all sample points in the new dataset have been obtained.

[0065] For example, the new dataset contains 100 suspected normal data points and 20 suspected abnormal data points, and these 20 suspected abnormal data points correspond to 3 suspected abnormal working condition clusters. By replacing these 20 suspected abnormal data points with the cluster centers of their respective clusters, the new dataset becomes 100 suspected normal data points and 3 suspected abnormal data points. These 3 suspected abnormal data points become extremely isolated and sparse, thereby improving the accuracy of identifying abnormal working conditions during the processing.

[0066] S5. Assign the reconstruction anomaly score of the cluster center of the suspected abnormal operating condition cluster to all data points within the cluster; input the original dataset into the isolated forest algorithm to obtain the original anomaly score of each data point; take the maximum value between the reconstruction anomaly score and the original anomaly score as the warning coefficient of each data point; and perform control and warning based on the warning coefficients of data points at all historical moments.

[0067] It should be noted that, considering that the sample points participating in the suspected abnormal operating condition clusters are represented by the reconstruction anomaly scores obtained from their cluster centers, and that the feature distribution of data points within the same cluster is similar, this means that the isolation of the reconstruction anomaly scores of the sample points participating in each suspected abnormal operating condition cluster in the constructed dataset represents the isolated performance of the data points within their respective clusters. Therefore, by combining the reconstruction anomaly scores of the cluster centers of each suspected abnormal operating condition cluster, the reconstruction anomaly scores of all data points within their respective clusters are obtained.

[0068] Specifically, the reconstruction anomaly score of the cluster center of each suspected abnormal operating condition cluster is obtained through iteration, and this score is then assigned to all data points within that cluster. This yields the reconstruction anomaly score of each data point at each time step after the new dataset is constructed.

[0069] Since the isolated forest algorithm is prone to misclassifying potentially abnormal data points as normal points in this scenario, meaning that the abnormal scores corresponding to potentially abnormal data points are relatively small after processing by the traditional algorithm, it is necessary to comprehensively consider the maximum abnormal score after processing by both algorithms as the warning coefficient to ensure accurate monitoring of potential abnormal states in milling machine processing.

[0070] Furthermore, the Isolation Forest algorithm is input into all data points of the original dataset to obtain the original anomaly score for each data point in the original dataset. It should be noted that the Isolation Forest algorithm for obtaining anomaly scores is a well-known technique and will not be elaborated here. In addition, in this embodiment of the invention, it is necessary to ensure that the number of isolation trees is consistent.

[0071] Calculate the warning coefficient for each data point in the original dataset. The warning coefficient for each data point is equal to the maximum of the original anomaly score for each data point in the original dataset and the reconstructed anomaly score for each data point after constructing the new dataset.

[0072] The system iterates through all historical data points during the current milling process, checking their warning coefficients. If the number of data points with warning coefficients exceeding a preset alarm threshold exceeds 10% of the total number of historical data points, an abnormality is considered to exist in the current milling process. Corresponding control program instructions are then generated and input into the control unit to issue a warning, notifying relevant personnel to optimize and adjust parameters. The specific value of the preset alarm threshold can be set according to the actual application scenario and requirements; in this invention, the alarm threshold is set to 0.7.

[0073] This completes the intelligent monitoring of milling machine processing.

[0074] This invention also discloses an intelligent monitoring system for milling machine machining, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, an intelligent monitoring method for milling machine machining according to the present invention is implemented.

[0075] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A method for intelligent monitoring of milling machine processing, characterized in that, include: Collect data values ​​for each dimension at all historical moments during the current milling process, obtain the data points for each historical moment, and construct the original dataset; Cluster all data points in the original dataset to obtain several clusters; determine the density of each cluster based on the distribution of distances between data points within each cluster, satisfying the following: ； For the first The density of each cluster; For the first Within the cluster, the first The data point and the Euclidean distance between data points; For the first The index value of data points within each cluster; For the first The total number of data points within each cluster; It is a natural exponential function; Based on density, select clusters with stable cutting states; Based on the trend changes between each data point within each cluster and the cluster centers of the stable cutting state clusters, the necessity of each cluster as part of the overall dataset is determined, satisfying the following: ； For the first The necessity of individual clusters as part of the overall dataset; For the first Within the cluster, the first Cluster centers of data points and stable cutting state clusters The slope values ​​between; To determine the absolute value sign; For standard normalization functions; Based on necessity, dynamic thresholds are calculated, and several clusters are divided into suspected abnormal operating condition clusters and suspected normal operating condition clusters. A new dataset is constructed by combining the cluster centers of all suspected abnormal operating condition clusters with the data points within all suspected normal operating condition clusters. The new dataset is input into the Isolation Forest algorithm to determine the reconstruction anomaly score for each sample point in the new dataset, satisfying the following: ； In the new dataset, the first Reconstruction anomaly score for each sample point; In the new dataset, the first The sample point and the first The total number of data points within the cluster of each sample point; The index value of the sample points in the new dataset; This represents the total number of sample points in the new dataset. In the new dataset, the maximum total number of data points within the cluster to which all sample points belong; In the new dataset, the first The sample point at the th th Path length in a single isolated tree; The index value and total number of the isolation tree; The reconstruction anomaly score of the cluster center of the suspected abnormal operating condition cluster is assigned to all data points within the cluster; the original dataset is input into the Isolation Forest algorithm to obtain the original anomaly score of each data point; The maximum value between the reconstructed anomaly score and the original anomaly score will be used as the warning coefficient for each data point. Based on the warning coefficients of all historical data points, control and warning are implemented, including: traversing the warning coefficients of all historical data points during the current milling process; if the number of data points with warning coefficients greater than the preset alarm threshold exceeds 10% of the total number of data points in all historical data points, it is determined that there is an abnormality in the current milling process, and control and warning are implemented.

2. The intelligent monitoring method for milling machine processing according to claim 1, characterized in that, The clustering of all data points in the original dataset is achieved through an iterative self-organizing clustering algorithm.

3. The intelligent monitoring method for milling machine processing according to claim 1, characterized in that, The step of filtering stable cutting state clusters based on density includes: Iterate through the density of each cluster and take the cluster corresponding to the maximum density as the stable cutting state cluster.

4. The intelligent monitoring method for milling machine processing according to claim 3, characterized in that, The process of calculating dynamic thresholds based on necessity and dividing several clusters into suspected abnormal operating condition clusters and suspected normal operating condition clusters includes: The necessity of each cluster as part of the overall dataset is traversed, and the dynamic threshold is equal to the average of the maximum and minimum values ​​of the necessity of each cluster as part of the overall dataset. Clusters whose necessity is greater than the dynamic threshold are marked as clusters suspected of abnormal operating conditions, and the remaining clusters are marked as clusters suspected of normal operating conditions.

5. The intelligent monitoring method for milling machine processing according to claim 1, characterized in that, The path length is obtained by inputting all sample points of the new dataset into the Isolation Forest algorithm to construct multiple isolation trees.

6. A milling machine machining intelligent monitoring system, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a milling machine intelligent monitoring method according to any one of claims 1-5.

Images