Laser cutting machine operating parameter monitoring method and system

Abstract

The present invention relates to the field of data processing technology, and more specifically to a method and system for monitoring the operating parameters of a laser cutting machine. The method comprises: collecting power data and cutting speed data during operation of the laser cutting machine, evaluating the authenticity of the power data based on the degree of outliers in the power data, and analyzing the degree of correlation between the power data and the corresponding cutting speed data; obtaining multiple nearest power data combinations of the power data to be analyzed, and the angle between each nearest power data combination; determining the optimal coefficient of each nearest power data combination based on the degree of correlation between the power data and the corresponding cutting speed, as well as the degree of authenticity of the power data; and calculating the anomaly score of the power data to be analyzed in combination with the angle and the optimal coefficient to perform anomaly monitoring. The method improves the accuracy of anomaly monitoring through multi-dimensional data association analysis and anomaly score calculation.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and more particularly to a method and system for monitoring operating parameters of a laser cutting machine. Background Art

[0002] With the rapid development of intelligent manufacturing and industry, the manufacturing industry faces new challenges in improving equipment automation and production efficiency. As core equipment in modern sheet metal processing, the operational stability of laser cutting machines has a direct impact on production efficiency and product quality. In particular, during the laser cutting process, abnormal fluctuations in laser power can lead to quality issues such as kerf burrs and material overburning. Therefore, real-time monitoring and anomaly detection of laser power data has become a crucial means of improving equipment reliability and production efficiency.

[0003] The existing FastABOD (Fast Angle-Based Outlier Detection) algorithm is an unsupervised data anomaly detection method that identifies outliers in the data by calculating the angular difference between a data point and its nearest neighbor. This method relies on the weighted variance of the geometric angles between a data point and its neighbors to calculate an anomaly score. A high anomaly score generally indicates that the data point behaves abnormally in the feature space.

[0004] However, laser cutting machines are often affected by electromagnetic interference during actual operation, especially when operating at high power or high speeds. Electromagnetic noise can distort power data. In this case, if the neighboring data combination selected in the FastABOD algorithm contains distorted data, the calculation of the anomaly score may be inaccurate, thereby reducing the reliability of data anomaly detection.

[0005] Therefore, there is an urgent need for a method that can accurately and reliably monitor the operating parameters of the laser cutting machine to avoid the influence of parameter distortion on the monitoring results. Summary of the Invention

[0006] In order to solve the problem that in the process of monitoring the operating parameters of a laser cutting machine, distorted power data may lead to inaccurate calculation of anomaly scores, thereby affecting the accuracy and reliability of anomaly monitoring, the present invention proposes a method and system for monitoring the operating parameters of a laser cutting machine.

[0007] In a first aspect, the present invention provides a method for monitoring operating parameters of a laser cutting machine, the method comprising:

[0008] Collect power data and cutting speed data during the operation of the laser cutting machine. The power data and cutting speed data are based on a one-to-one correspondence with the collection time;

[0009] Determine the degree of outlier of each power data according to the Euclidean distance between each power data and the historical power data, and then evaluate the authenticity of the power data according to the degree of outlier; determine the correlation between each power data and the corresponding cutting speed data through correlation analysis;

[0010] Taking any power data as the power data to be analyzed, obtaining multiple nearest power data combinations of the power data to be analyzed, and the angle value between the power data to be analyzed and each nearest power data combination;

[0011] For each power data included in each nearest power data combination, a preferred coefficient of each nearest power data combination is determined based on a weighted result of the authenticity of the power data according to the correlation between the power data and the corresponding cutting speed data;

[0012] The anomaly score of the power data to be analyzed is determined based on the angle value between the power data to be analyzed and each nearest power data combination, as well as the preferred coefficient of each nearest power data combination. The anomaly score is used to monitor whether the operating parameters are abnormal.

[0013] This technical solution ensures accurate correspondence in the time dimension by synchronously collecting power and cutting speed data. The data authenticity is quantified based on the degree of outliers in the power data, and the possibility of data distortion is preliminarily analyzed. By analyzing the correlation between the power data and the cutting speed data, the distorted data is further locked. By obtaining the nearest power data combination and angle value, the relative position of the power data to be analyzed and the nearest power data combination can be reflected, thereby revealing whether they are within a reasonable range of physical changes. Through angle calculation, abnormal fluctuations can be found in the geometric characteristics of the data, which helps to eliminate normal fluctuations caused by external interference or operational changes, and enhance the robustness of the detection method. The optimal coefficient of the nearest power data combination is determined by the authenticity of the power data and the correlation between the power data and the corresponding cutting speed data. The optimal coefficient and angle calculation are combined to evaluate the abnormal score of the power data to be analyzed, making the abnormal score more accurate. This means that monitoring not only considers the degree of outliers and correlation, but also introduces multi-level information to confirm whether the data deviates from the normal working state, thereby improving the accuracy and reliability of monitoring.

[0014] Preferably, the abnormality score of the power data to be analyzed is determined based on the following method: using the preferred coefficient of each nearest laser power combination of the power data to be analyzed as a weight, performing a weighted average operation on the angle value between the power data to be analyzed and each nearest power data combination to obtain a weighted average angle value; using the preferred coefficient of each nearest laser power combination of the power data to be analyzed as a weight, combining the weighted average angle value to calculate the weighted variance, and using the weighted variance as the abnormality score of the power data to be analyzed.

[0015] This technical solution achieves accurate quantification of the abnormal characteristics of the power data to be analyzed by incorporating the preferred coefficient as a weight into the weighted average and variance calculation of the angle value. It not only ensures sensitivity to real anomalies but also enhances robustness to normal fluctuations, achieving a dual improvement in the accuracy and reliability of anomaly detection.

[0016] Preferably, the method for monitoring whether an operating parameter is abnormal through anomaly scores includes: obtaining the anomaly scores of all historical power data of the power data to be analyzed, and determining an anomaly score threshold based on the anomaly scores of all historical power data using a quantile method; if the anomaly score of the power data to be analyzed is greater than the anomaly score threshold, determining that the power data to be analyzed is abnormal power data; if the anomaly score of the power data to be analyzed is not greater than the anomaly score threshold, determining that the power data to be analyzed is normal power data.

[0017] Preferably, the preferred coefficient of each nearest neighbor power data combination is determined based on the following method: for each power data contained in each nearest neighbor power data combination, the authenticity of the power data is weighted using the correlation between the power data and the corresponding cutting speed data, and the weighted result is used as the preferred coefficient of the power data; the average of the preferred coefficients of all power data contained in each nearest neighbor power data combination is used as the preferred coefficient of the nearest neighbor power data combination.

[0018] This technical solution achieves multi-level screening of data credibility and enhancement of combination reliability by weightedly fusing the authenticity of power data with the correlation of corresponding cutting speed data, and then using the average of the preferred coefficients of each power data in the combination as the preferred coefficient of the nearest power data combination.

[0019] Preferably, the degree of outlier is determined based on the following method: constructing a coordinate system with the acquisition time as the horizontal axis and the power data as the vertical axis; each power data and its historical power data correspond to a point in the coordinate system; calculating the Euclidean distance between each power data and each of its historical power data based on the coordinate system, and taking the sum of the Euclidean distances between the power data and all the historical power data as the degree of outlier of the power data.

[0020] Preferably, the method for obtaining a plurality of nearest neighbor power data combinations of the power data to be analyzed includes:

[0021] The point of the power data to be analyzed in the coordinate system is the target point, and the point of each historical power data of the power data point to be analyzed in the coordinate system is the candidate point; the Euclidean distance between the target point and each candidate point is calculated, and the K candidate points with the closest Euclidean distance are selected. The K candidate points are combined in pairs without duplication to obtain A combination, The historical power data combination corresponding to the combinations is used as the nearest power data combination of the power data to be analyzed, where K is a preset positive integer. It is the permutation and combination symbol.

[0022] Preferably, the method for obtaining the angle value between the power data to be analyzed and each nearest neighbor power data combination includes: constructing two vectors with the two power data contained in each nearest neighbor power data combination and the power data to be analyzed, the directions of the two vectors are respectively pointed from the power data to be analyzed to the two power data, and the sizes of the vectors are respectively the Euclidean distances between the power data to be analyzed and the two power data; and taking the angle between the two vectors as the angle value between the power data to be analyzed and the nearest neighbor power data combination.

[0023] Preferably, evaluating the authenticity of the power data through the degree of outliers is based on the following method: taking the degree of outliers of each power data as a negative indicator for evaluating the authenticity of the power data, evaluating the confidence of the outlier degree by the difference between the degree of outliers of each power data and the maximum value of the outlier degrees of all historical power data; and comprehensively determining the authenticity of the power data based on the degree of outliers and its confidence.

[0024] This technical solution uses the outlier degree of power data as a negative indicator of authenticity, and combines the difference between the outlier degree and the historical maximum outlier degree to evaluate the confidence level. It constructs a multi-level authenticity assessment system that can accurately quantify the authenticity of power data. The accurate authenticity reduces the possibility of misjudgment of anomalies due to data distortion during anomaly monitoring.

[0025] Preferably, determining the correlation between each power data and the corresponding cutting speed data through correlation analysis is based on the following method: calculating the first Pearson correlation coefficient between the power data and the cutting speed data based on each power data and the corresponding cutting speed data, as well as all cutting speed data corresponding to all historical power data; calculating the second Pearson correlation coefficient between the power data and the cutting speed data based only on all cutting speed data corresponding to all historical power data of each power data; and taking the difference between the first Pearson coefficient and the second Pearson coefficient as the correlation between the power data and the corresponding cutting speed data.

[0026] This technical solution constructs a quantitative indicator that reflects the correlation between power data and cutting speed data. The first Pearson coefficient comprehensively considers the correlation between the current power data and its corresponding cutting speed and all historical data. The second Pearson coefficient is based only on the correlation pattern of historical data. The difference between the two can keenly capture the deviation between the current data and the historical pattern. This difference analysis mechanism, by converting the physical process correlation pattern into a quantitative difference indicator, provides a reliable logical constraint for the subsequent calculation of anomaly scores and enhances the ability to distinguish distorted data such as electromagnetic interference.

[0027] In a second aspect, the present invention proposes a laser cutting machine operating parameter monitoring system, which includes a memory and a processor, wherein a computer program is stored in the memory, and the processor executes the computer program to implement the steps of any one of the operating parameter monitoring methods.

[0028] The present invention has the following effects:

[0029] By combining the physical correlation between power data and cutting speed data, the present invention proposes an anomaly detection method based on the optimal coefficient. By analyzing the authenticity of the power data, the influence of the anomaly score of the distorted data caused by electromagnetic interference is avoided, making the anomaly score of the power data more accurate, thereby improving the accuracy and reliability of the laser cutting machine operating parameter monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic flow chart of the method of the present invention. DETAILED DESCRIPTION

[0031] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention.

[0032] The present invention provides a method and system for monitoring operating parameters of a laser cutting machine, such as Figure 1 As shown in , including:

[0033] S1: Collect laser power data and cutting speed data when the laser cutting machine is running.

[0034] During the operation of the laser cutting machine, a laser power meter and laser velocimeter are used to synchronously collect the power and cutting speed data of the laser cutting machine in real time. This means that a power value and a cutting speed value are obtained at each acquisition moment. If the acquisition frequency is set to 1kHz, 1000 power and cutting speed data points are collected per second.

[0035] Through this step, the power data and cutting speed data of the laser cutting machine during operation can be accurately and synchronously collected, and a high-quality data foundation can be provided for subsequent data processing and analysis.

[0036] S2: Determine the true degree of the power data according to the outlier degree of the power data.

[0037] In the operation monitoring of laser cutting machines, it is crucial to ensure the authenticity of their operating parameters. These parameter data reflect the working status of the machine. However, in the actual collection process, some data may be distorted due to factors such as electromagnetic interference, thereby affecting the accuracy of judgment.

[0038] In order to accurately and reliably monitor the power data of the laser cutting machine, this step calculates the authenticity of the power data. The size of the authenticity reflects the possibility that the power data is caused by electromagnetic interference. The smaller the authenticity, the more likely it is caused by electromagnetic interference.

[0039] Specifically, by analyzing the degree of deviation of each power data point from its historical power data, we can determine the degree of outlier. The greater the degree of outlier, the further the power data point deviates from the historical level, which means that the data point is more likely to be distorted data caused by electromagnetic interference, and the less true it is, and vice versa. This method not only effectively determines whether power data is affected by electromagnetic interference, but also improves data accuracy, ensuring the scientific and effective nature of subsequent analysis and decision-making.

[0040] In one embodiment, the outlier level of each power data is determined as follows:

[0041] For each power data, the 50 (empirical value, adjustable) power data before the power data are taken as all the historical power data of the power data.

[0042] A coordinate system is constructed with the acquisition time as the horizontal axis and the power data as the vertical axis. Then, each power data and its historical power data corresponds to a point in the coordinate system.

[0043] Based on this coordinate system, the Euclidean distance between each power data point and each historical power data point is calculated, resulting in a total of 50 Euclidean distances. The sum of these 50 Euclidean distances is used as the outlier degree of the power data point. The larger the sum of the Euclidean distances, the farther the power data point is from its historical power data point, and the greater the outlier degree, and vice versa.

[0044] Similarly, for each historical power data of the power data, 50 power data are acquired forward based on each historical power data to obtain all historical power data of each historical power data, and the outlier degree of each historical power data of the power data is calculated using the same method.

[0045] For example, for the 100th power data, the 50th power data to the 99th power data are taken as all the historical power data of the 100th power data. According to the coordinate system, the Euclidean distances between the 50th power data to the 99th power data and the 100th power data are calculated in sequence, and a total of 50 Euclidean distances are obtained. The sum of these 50 Euclidean distances is taken as the outlier degree of the 100th power data. The 99th power data is the first historical power data of the 100th power data, and the 98th power data is the second historical power data of the 100th power data. Then 49 to 98 are the historical power data of the first historical power data, and 48 to 97 are the historical power data of the second historical power data. Similarly, the Euclidean distance is calculated based on the coordinate system to obtain the outlier degree of the first historical power data and the outlier degree of the second historical power data.

[0046] In one embodiment, after obtaining the outlier degree of each power data and each historical power data, the authenticity of the power data is comprehensively determined in combination with the confidence level of the outlier degree, specifically including:

[0047] The outlier degree of each power data is used as a negative indicator to evaluate the true degree of the power data, that is, the greater the outlier degree, the smaller the true degree.

[0048] The maximum value of the outlier degree of all historical power data of the power data is obtained, and the confidence level of the outlier degree of the power data is evaluated by the difference between the outlier degree of the power data and the maximum value. If the difference is smaller, it means that the outlier degree of the power data is closer to the maximum outlier degree of the historical power data, reflecting that the more the power data deviates from the historical power data, the greater the confidence level of the outlier degree of the power data is increased. If the difference is larger, it means that the outlier degree of the power data is further away from the maximum outlier degree of the historical power data, reflecting that the closer the power data is to the historical power data, the smaller the confidence level of the outlier degree of the power data is increased.

[0049] Therefore, for each power data, its authenticity can be calculated by the following formula:

[0050]

[0051] In this formula, Indicates the authenticity of the power data. Indicates the degree of outlier of the power data, Indicates the maximum outlier degree of all historical power data of this power data. and is the preset hyperparameter, , in order to prevent The value is 0. To prevent zero parameters, , here is to prevent A value of 0 causes subsequent numerical calculation results to be unstable. is the absolute value symbol, is the natural exponential function, is a normalization operation, so that The value of Within the range.

[0052] In this formula, The larger the value is, the greater the degree of outlier of the power data is, the more the power data deviates from the historical level, the more likely the power data is generated by electromagnetic interference, the greater the possibility of data distortion, and the smaller the true degree is, and vice versa. Construct a nonlinear negative correlation between the degree of outliers and the true degree. When the degree of outliers increases, The structure decays exponentially, which is consistent with the characteristic that the greater the degree of outlier, the faster the degree of truth decreases.

[0053] In this formula, The smaller it is, the closer the outlier degree of the laser power data is to the maximum outlier degree, which means that the more the power data deviates from the historical level, the greater the credibility is, and the greater the possibility that the power data is distorted data, so the true degree will be smaller, and vice versa.

[0054] In short, the sensitivity to changes in the degree of outliers is enhanced through exponential decay, and the ability to distinguish interference patterns is further improved with the help of confidence analysis. Ultimately, the assessment of the degree of authenticity not only conforms to the logic that the more stable the power data, the more reliable it is, but also can dynamically adapt to the complex scenarios of electromagnetic interference in the laser cutting process.

[0055] S3: Determine the correlation between the power data and the corresponding cutting speed data through correlation analysis.

[0056] When monitoring laser cutting machine operation, power and cutting speed are two core performance indicators. Their correlation directly impacts the machine's operating efficiency and cutting quality. To more accurately understand and predict the machine's operating status, this step analyzes the correlation between power and cutting speed data to reveal how these two variables change under different operating conditions and how they interact.

[0057] Under normal operating conditions, power and cutting speed data should exhibit a certain degree of synergistic coupling. Power determines the material's melting capacity, while cutting speed affects heat input. Both must adhere to the principle of constant energy density. For example, when cutting carbon steel, for every 100W increase in power, the cutting speed typically increases by 5-8mm / s. However, when subject to electromagnetic interference, the physical correlation between power and cutting speed is disrupted, resulting in an artificially high power output and a failure to increase the actual cutting speed according to process rules.

[0058] In one embodiment, the correlation between each power data and the corresponding cutting speed data is determined as follows:

[0059] The Pearson correlation coefficient is the most commonly used method to measure the linear correlation between two variables. It can clearly reveal the degree of correlation between power data and cutting speed data. Using this statistical method can effectively avoid subjective judgment and provide a quantitative and objective indicator.

[0060] First, each power data and all its historical power data constitute a power data sequence, and the cutting speed data corresponding to the power data and all the cutting speed data corresponding to all the historical power data of the power data constitute a cutting speed data sequence. The Pearson correlation coefficient between the power data and the cutting speed data is calculated based on the power data sequence and the cutting speed data sequence. , It reflects the real-time correlation mode of power and cutting speed under the current working conditions at the corresponding moment of the power data.

[0061] Then, a power data sequence is constructed based on all the historical power data of each power data, and a cutting speed data sequence is constructed based on all the cutting speed data corresponding to all the historical power data of the power data. The Pearson correlation coefficient between the power data and the cutting speed data is calculated based on the power data sequence and the cutting speed data sequence. , This represents the baseline correlation mode between power and cutting speed when the equipment is operating normally.

[0062] Finally, As the correlation between the power data and its corresponding cutting speed data. Reflects the degree of deviation between the correlation mode under the current working condition and the benchmark correlation mode. The larger the value is, the more the current working conditions deviate from the historical process logic, and there may be distortion or abnormality.

[0063] In summary, the double correlation coefficient difference method is used to transform the physical coupling law of power and cutting speed in the laser cutting process into a quantifiable abnormality indicator. This method not only conforms to the rigorous benchmark comparison method in statistics, but also adapts to the need to distinguish between dynamic changes in process parameters and data distortion in industrial scenarios.

[0064] S4: Selecting power data to be analyzed, obtaining a plurality of nearest power data combinations of the power data to be analyzed, and an angle value between the power data to be analyzed and the nearest power data combinations.

[0065] This step is based on the idea of ​​the FastABOD algorithm to construct the nearest neighbor power data combination and angle calculation model in the space-time coordinate system, thereby achieving accurate characterization of the local geometric features of the power data.

[0066] The multiple nearest neighbor power data combinations of the power data to be analyzed are obtained according to the following method:

[0067] The point corresponding to the power data to be analyzed in the coordinate system (constructed in step S2) is used as the target point, and the point of each historical power data of the power data point to be analyzed in the coordinate system is used as the candidate point;

[0068] Calculate the Euclidean distance between the target point and each candidate point, select the K candidate points with the closest Euclidean distance, and combine the K candidate points in pairs without duplication to obtain A combination, The historical power data combinations corresponding to the combinations are taken as the nearest power data combinations of the power data to be analyzed, where K is a preset positive integer and the empirical value is 10, so a total of 45 nearest power data combinations are obtained. It is the permutation and combination symbol.

[0069] The angle value between the power data to be analyzed and each combination of the nearest power data is obtained according to the following method:

[0070] Two vectors are constructed by combining the two power data contained in each nearest neighbor power data combination with the power data to be analyzed. The directions of the two vectors are respectively directed from the power data to be analyzed to the two power data. The sizes of the vectors are the Euclidean distances between the power data to be analyzed and the two power data. The angle between the two vectors is taken as the angle value between the power data to be analyzed and the nearest neighbor power data combination.

[0071] Under normal operating conditions, power data should follow the principle of constant energy density, and its distribution in the coordinate system usually shows geometric clustering, that is, the relative position changes between data points are small, and the angle values ​​are usually concentrated in a small range (for example, the mean is usually less than 30°). This geometric clustering reflects the stability and consistency of the equipment under normal operating conditions.

[0072] When abnormal interference occurs in equipment, the power data deviates from its normal pattern, resulting in a significant shift in its distribution within the coordinate system. In this case, the angle between the power data being analyzed and its nearest neighbor increases, and the dispersion of the angles also increases significantly. This change indicates a change in the equipment or process state, providing an effective indicator for anomaly detection.

[0073] S5: determining a preferred coefficient of the nearest power data combination according to the correlation between the power data included in the nearest power data combination and the corresponding cutting speed data, and the authenticity of the power data.

[0074] During the laser cutting process, electromagnetic interference may cause power data distortion. If the equal-weighted neighbor combination analysis of the traditional FastABOD algorithm is directly used, the erroneous participation of the distorted data will lead to misjudgment of the anomaly score, affecting the accuracy of the anomaly monitoring results. For example, when a nearest neighbor power data combination contains distorted data, the anomaly score will be too high or too low.

[0075] Therefore, this step determines the optimal coefficient of each nearest neighbor power data combination by fusing the authenticity of the power data (reflecting the physical possibility of the data deviating from the historical pattern) and the correlation between the power data and the corresponding cutting speed data (reflecting the consistency of the process logic).

[0076] In one embodiment, the preferred coefficient of the nearest neighbor power data combination is determined based on the following method:

[0077] For the power data to be analyzed, obtain the first The nearest neighbor power data combination, The nearest neighbor power data combination contains two power data, which are recorded as and .

[0078] for ,use The correlation between the corresponding cutting speed data The weighted operation is performed on the authenticity of Specifically, the following relationship is satisfied:

[0079]

[0080] In this formula, for The preferred coefficient, for The degree of authenticity, for The correlation between the corresponding cutting speed data, To prevent zero parameters, , is the normalization function. This is to prevent the correlation after normalization is 0, as can be seen from step S3, the correlation It is the difference between two Pearson correlation coefficients, and the normalized value is in the range [0,1].

[0081] for ,use The correlation between the corresponding cutting speed data The weighted operation is performed on the authenticity of Specifically, the following relationship is satisfied:

[0082]

[0083] In this formula, for The preferred coefficient, for The degree of authenticity, for The correlation between the corresponding cutting speed data, To prevent zero parameters, , is the normalization function. This is to prevent the correlation after normalization is 0, as can be seen from step S3, the correlation It is the difference between two Pearson correlation coefficients, and the normalized value is in the range [0,1].

[0084] Next, and The mean of the power data to be analyzed is The optimal coefficient of the nearest neighbor power data combination.

[0085] In summary, the automatic filtering of distorted data is achieved through the optimization coefficient, so that the optimization coefficients of the nearest power data combinations with high authenticity and strong correlation are larger, and thus they are given greater weights when participating in the calculation of anomaly scores, and vice versa, they are given smaller weights, thus avoiding the interference of distorted data on anomaly monitoring results.

[0086] S6: Determine an anomaly score of the power data to be analyzed based on the angle value between the power data to be analyzed and the combination of the nearest neighboring power data, and the optimal coefficient of the combination of the nearest neighboring power data.

[0087] In one embodiment, the anomaly score of the power data to be analyzed is determined based on the following method:

[0088] First, using the preferred coefficient of each nearest laser power combination of the power data to be analyzed as a weight, a weighted average operation is performed on the angle value between the power data to be analyzed and each nearest power data combination to obtain a weighted average angle value.

[0089] Specifically, the weighted average operation is calculated according to the following formula:

[0090]

[0091] In this formula, The weighted average angle value represents the angle value between the power data to be analyzed and all its nearest power data combinations. Represents the total number of all nearest neighbor power data combinations of the power data to be analyzed, Indicates the first The optimal coefficient of the nearest neighbor power data combination, Indicates the power data to be analyzed and The angle value between the nearest power data combinations.

[0092] In summary, since the optimization coefficient comprehensively considers the authenticity of the power data and its correlation with the cutting speed, more reliable adjacent power combinations will obtain higher weights when calculating the weighted average angle value, thereby avoiding the misleading of the overall trend by distorted data.

[0093] Next, the optimal coefficient of each nearest neighbor laser power combination of the power data to be analyzed is used as a weight, and the weighted variance is calculated in combination with the weighted average angle value, and the weighted variance is used as the abnormal score of the power data to be analyzed. Specifically, the weighted variance is calculated according to the following formula:

[0094]

[0095] In this formula, is the weighted variance, which represents the abnormality score of the power data to be analyzed, The weighted average angle value represents the angle value between the power data to be analyzed and all its nearest power data combinations. Indicates the power data to be analyzed and The angle value between the nearest power data combinations, Represents the total number of all nearest neighbor power data combinations of the power data to be analyzed, Indicates the first The optimal coefficient of the nearest neighbor power data combination. The larger the value, the closer the power data to be analyzed is to the The more the angle value between the nearest neighboring power data combinations deviates from the average level, the greater the possibility that the power data to be analyzed is abnormal, and vice versa.

[0096] Using weighted variance as the anomaly score essentially constructs a two-dimensional evaluation system that multiplies geometric deviation by data credibility. The geometric anomaly of the power data is quantified by the difference between the angle and the weighted average angle. The optimal coefficient is used to filter noise and strengthen process constraints, so that the anomaly score not only reflects the physical characteristics of the data distribution, but also incorporates the process logic of laser cutting.

[0097] In summary, the weighted variance calculated based on the optimization coefficient can amplify the angular dispersion between the abnormal data points and the adjacent combinations. By quantifying the variance weighted by the credibility, an accurate abnormality assessment can be performed on the power data to be analyzed.

[0098] S7: Monitor whether the operating parameters of the laser cutting machine are abnormal through the abnormality score.

[0099] Obtain the anomaly scores of 20 sets (1000 sets) of historical power data (empirical values) before the power data to be analyzed. The method for obtaining the anomaly score of each historical power data is the same as the method for obtaining the anomaly score of the power data to be analyzed.

[0100] From a statistical perspective, 20 sets of data (1,000 samples) can usually significantly reduce the estimation error of the 99% quantile threshold, meeting the sample size requirements of the law of large numbers, ensuring that the threshold can accurately characterize the distribution of normal data, and avoiding problems such as threshold offset and poor adaptability to multiple working conditions caused by a single set of data being limited to a single process scenario.

[0101] The anomaly scores of these historical power data are sorted from smallest to largest. The sorted anomaly scores are divided according to the 100th percentile, and the anomaly data is divided into 100 intervals. Based on the quantile division, the 99th percentile is selected as the anomaly score threshold. If the anomaly score of the power data to be analyzed is greater than the anomaly score threshold, the power data to be analyzed is determined to be abnormal, and technical personnel are promptly notified to perform maintenance on the laser cutting machine. If the anomaly score of the power data to be analyzed is less than or equal to the anomaly score threshold, the power data to be analyzed is determined to be normal power data, and the laser cutting machine maintains normal operation.

[0102] Through this step, it is possible to accurately identify whether the operating parameters of the laser cutting machine are abnormal based on the comparison result of the abnormality score and the abnormality score threshold, thereby effectively monitoring its operating status and helping to promptly discover operating faults of the laser cutting machine.

[0103] The present invention also proposes a laser cutting machine operating parameter monitoring system, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the operating parameter monitoring method.

Claims

1. A method for monitoring operating parameters of a laser cutting machine, characterized in that: include: Collect power data and cutting speed data during the operation of the laser cutting machine. The power data and cutting speed data are based on a one-to-one correspondence with the collection time; Determine the degree of outlier of each power data according to the Euclidean distance between each power data and the historical power data, and then evaluate the authenticity of the power data according to the degree of outlier; determine the correlation between each power data and the corresponding cutting speed data through correlation analysis; Taking any power data as the power data to be analyzed, obtaining multiple nearest power data combinations of the power data to be analyzed, and the angle value between the power data to be analyzed and each nearest power data combination; For each power data included in each nearest power data combination, a preferred coefficient of each nearest power data combination is determined based on a weighted result of the authenticity of the power data according to the correlation between the power data and the corresponding cutting speed data; The anomaly score of the power data to be analyzed is determined based on the angle value between the power data to be analyzed and each nearest power data combination, as well as the preferred coefficient of each nearest power data combination. The anomaly score is used to monitor whether the operating parameters are abnormal.

2. The laser cutting machine operating parameter monitoring method according to claim 1, characterized in that: The anomaly score of the power data to be analyzed is determined based on the following: Using the optimal coefficient of each nearest laser power combination of the power data to be analyzed as a weight, a weighted average operation is performed on the angle value between the power data to be analyzed and each nearest power data combination to obtain a weighted average angle value; The preferred coefficient of each nearest laser power combination of the power data to be analyzed is used as a weight, and the weighted variance is calculated in combination with the weighted average angle value, and the weighted variance is used as the abnormal score of the power data to be analyzed.

3. The laser cutting machine operating parameter monitoring method according to claim 1, characterized in that: Methods for monitoring whether operating parameters are abnormal by using anomaly scores include: Obtain the anomaly scores of all historical power data of the power data to be analyzed, and use the quantile method to determine the anomaly score threshold based on the anomaly scores of all historical power data; if the anomaly score of the power data to be analyzed is greater than the anomaly score threshold, determine that the power data to be analyzed is abnormal power data; if the anomaly score of the power data to be analyzed is not greater than the anomaly score threshold, determine that the power data to be analyzed is normal power data.

4. The laser cutting machine operating parameter monitoring method according to claim 1, characterized in that: The preferred coefficient of each nearest neighbor power data combination is determined based on the following method: For each power data included in each nearest power data combination, the authenticity of the power data is weighted using the correlation between the power data and the corresponding cutting speed data, and the weighted result is used as the optimization coefficient of the power data; The average value of the preferred coefficients of all power data included in each nearest neighbor power data combination is used as the preferred coefficient of the nearest neighbor power data combination.

5. The method for monitoring operating parameters of a laser cutting machine according to claim 1, wherein: The degree of outlier is determined based on: A coordinate system is constructed with the acquisition time as the horizontal axis and the power data as the vertical axis. Each power data and its historical power data corresponds to a point in the coordinate system. The Euclidean distance between each power data and each historical power data is calculated based on the coordinate system, and the sum of the Euclidean distances between the power data and all the historical power data is used as the outlier degree of the power data.

6. The method for monitoring operating parameters of a laser cutting machine according to claim 5, characterized in that: The method for obtaining multiple nearest neighbor power data combinations of the power data to be analyzed includes: The point of the power data to be analyzed in the coordinate system is taken as the target point, and the point of each historical power data of the power data point to be analyzed in the coordinate system is taken as the candidate point; Calculate the Euclidean distance between the target point and each candidate point, select the K candidate points with the closest Euclidean distance, and combine the K candidate points in pairs without duplication to obtain A combination, The historical power data combination corresponding to the combinations is used as the nearest power data combination of the power data to be analyzed, where K is a preset positive integer. It is the permutation and combination symbol.

7. The method for monitoring operating parameters of a laser cutting machine according to claim 1, characterized in that: The method for obtaining the angle value between the power data to be analyzed and each combination of the nearest power data includes: Construct two vectors from the two power data contained in each nearest neighbor power data combination and the power data to be analyzed. The directions of the two vectors are from the power data to be analyzed to the two power data respectively. The sizes of the vectors are the Euclidean distances between the power data to be analyzed and the two power data respectively. The angle between the two vectors is used as the angle value between the power data to be analyzed and the combination of the nearest power data.

8. The method for monitoring operating parameters of a laser cutting machine according to claim 1, characterized in that: The evaluation of the authenticity of the power data by outlier level is based on the following method: The outlier degree of each power data is used as a negative indicator to evaluate the authenticity of the power data. The confidence level of the outlier degree is evaluated by the difference between the outlier degree of each power data and the maximum outlier degree of all historical power data. The authenticity of the power data is comprehensively determined based on the outlier degree and its confidence level.

9. The method for monitoring operating parameters of a laser cutting machine according to claim 1, wherein: The correlation analysis to determine the correlation between each power data and the corresponding cutting speed data is based on the following method: According to each power data and the corresponding cutting speed data, as well as all the cutting speed data corresponding to all the historical power data, the first Pearson correlation coefficient between the power data and the cutting speed data is calculated; based only on all the cutting speed data corresponding to all the historical power data of each power data, the second Pearson correlation coefficient between the power data and the cutting speed data is calculated; and the difference between the first Pearson coefficient and the second Pearson coefficient is used as the correlation between the power data and the corresponding cutting speed data.

10. A laser cutting machine operating parameter monitoring system, characterized in that: The operating parameter monitoring system includes a memory and a processor, the memory stores a computer program, and the processor executes the computer program to implement the steps of the operating parameter monitoring method according to any one of claims 1 to 9.

Images