Monitoring abnormal signal identification method and system in laser welding process

Abstract

The invention belongs to the technical field of laser welding, and particularly relates to a method and system for identifying monitoring abnormal signals in the laser welding process, and the method comprises the following steps: S1, obtaining an optical signal data template library in the laser welding process; s2, multi-channel local features are extracted from the data template library by sliding a time window; s3, detecting and eliminating abnormal data according to an isolated forest algorithm and the local features; according to the method, the accuracy and the reliability of the welding process monitoring system in the aspect of quality detection are enhanced by improving the recognition and elimination precision of the abnormal signals in the template library, the abnormal signals possibly influencing the welding quality can be recognized and eliminated more effectively, and the stability of the welding process and the quality of final products are ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of laser welding, and in particular relates to a method and system for identifying abnormal signals during laser welding process monitoring. Background Art

[0002] Laser welding is widely used in many fields such as lithium battery, automobile, aerospace and electronics due to its high efficiency and precision. In this process, real-time monitoring of welding quality is crucial to ensure the quality of the final product.

[0003] For example, the patent document with publication number CN116718595A discloses a process monitoring system, which builds a template library based on normal welding signals by collecting and analyzing various optical signals generated during the welding process, such as visible light, infrared light, and reflected light. Then, the system evaluates the welding quality by comparing the real-time detected signals with the data in the template library. However, in the process of establishing the template library, due to the inevitable existence of noise and abnormal signals in the welding process, these signals may originate from abnormalities of the weldment, equipment failures, or operational errors, resulting in deviations between the template library and the ideal normal state. This not only affects the accuracy of the template library, but also reduces the reliability of subsequent welding quality assessments.

[0004] At present, the method commonly used in the industry is to use the statistical characteristics of the overall signal to identify and remove abnormal signals by manually or adaptively setting thresholds. However, this method has some obvious limitations, including low accuracy of abnormal detection and difficulty in identifying local abnormalities. The accuracy of traditional abnormal signal identification methods is insufficient, resulting in errors in the monitoring system when identifying abnormal signals.

[0005] Based on this, it is necessary to improve the defects of the prior art to overcome the deficiencies in practical applications. Summary of the invention

[0006] Based on the above-mentioned shortcomings and deficiencies in the prior art, one of the objects of the present invention is to at least solve one or more of the above-mentioned problems in the prior art. In other words, one of the objects of the present invention is to provide a laser welding process monitoring abnormal signal identification method and system that meets one or more of the above-mentioned needs.

[0007] In order to achieve the above-mentioned object of the invention, the present invention adopts the following technical solutions:

[0008] The present invention provides a method for identifying abnormal signals during laser welding process monitoring, comprising the following steps:

[0009] S1, obtaining a light signal data template library for the laser welding process;

[0010] S2, extracting multi-channel local features from the data template library through a sliding time window;

[0011] S3. Detect and remove abnormal data based on the isolation forest algorithm and local features.

[0012] As a preferred solution, step S1 includes:

[0013] S11, collecting optical signals during welding, where the optical signals include visible light, infrared light and reflected light;

[0014] S12, performing data preprocessing on the collected welding multi-dimensional optical signal;

[0015] S13. Continuously collect normal welding data to build a template library of data to be identified.

[0016] As a preferred solution, step S12 includes:

[0017] S121, performing equal-length interception on all three-channel data of visible light, infrared light and reflected light collected in the template library to intercept valid data in the actual welding process;

[0018] S122, filtering and standardizing the intercepted equal-length light signal data, and stacking the three-channel light signals as a single welding sample signal.

[0019] As a preferred solution, step S2 includes:

[0020] S21, for the optical signal in the constructed template library, sequence segmentation is performed through a sliding time window;

[0021] S22, performing feature extraction on each time window signal after segmentation;

[0022] S23, repeat the above steps cyclically for all sample signals in the template library, and construct a sample library for each time window.

[0023] As a preferred solution, step S3 includes:

[0024] S31, by splitting the sample library in each time window multiple times into random sample sets, so as to construct an isolation forest corresponding to the time window;

[0025] S32, calculating the average path length of each sample on all isolated trees in the isolation forest to determine the local anomaly score;

[0026] S33, repeating the above steps for the sample libraries in all time windows to obtain the local anomaly score of each template library data in all time windows and fusing them to obtain the overall anomaly score of all template library data;

[0027] S34. Construct a decision function based on the overall anomaly score and the contamination rate, calculate the threshold and judge abnormal data or normal data to eliminate the abnormal data.

[0028] As a preferred solution, step S31 includes:

[0029] S311, randomly selecting a certain number of samples from the sample library to form a sample set;

[0030] S312, randomly selecting a feature from the sample set and determining a split value to construct an isolated tree;

[0031] S313, repeat the above steps to construct multiple isolated trees to form an isolation forest.

[0032] The present invention also provides a laser welding process monitoring abnormal signal recognition system, which is applied to the method described in any of the above schemes, comprising:

[0033] An acquisition module is used to obtain a template library of optical signal data of a laser welding process;

[0034] A feature extraction module is used to extract multi-channel local features from a data template library through a sliding time window;

[0035] The detection and elimination module is used to detect and eliminate abnormal data based on the isolation forest algorithm and local features.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] The present invention provides a method for identifying abnormal signals in laser welding process monitoring, which improves the recognition and elimination accuracy of abnormal signals in a template library, thereby enhancing the accuracy and reliability of the welding process monitoring system in quality detection, and is conducive to more effectively identifying and eliminating abnormal signals that may affect the welding quality, thereby ensuring the stability of the welding process and the quality of the final product. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, drawings of other embodiments can be obtained based on these drawings without paying any creative work.

[0039] Figure 1 is a flow chart of a method for identifying abnormal signals during monitoring of a laser welding process according to an embodiment of the present invention;

[0040] Figure 2is a schematic diagram of the flow chart of the local isolation forest algorithm according to an embodiment of the present invention;

[0041] Figure 3 is a schematic diagram of sample number display and abnormal data indication according to an embodiment of the present invention;

[0042] Figure 4 Schematic diagram of anomaly score results and comparison method results according to an embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to more clearly illustrate the embodiments of the present application, the specific implementation methods of the present application will be described below with reference to the accompanying drawings. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings and other implementation methods can be obtained based on these drawings without creative work.

[0044] In the description of the embodiments of the present application, the terms "upper", "lower", "front", "rear", etc., are based on the directions or positions shown in the drawings, and are only for the convenience of description and simplification of operation, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as a limitation of the present application. In addition, the terms "first", "second", etc. are only used to distinguish in the description and have no special meaning.

[0045] According to some embodiments of the present application, please refer to Figure 1 to Figure 2 As shown, a method for identifying abnormal signals during laser welding process monitoring is now provided, comprising the following steps:

[0046] S1. Obtain a template library of optical signal data of the laser welding process.

[0047] Specifically, the step S1 includes:

[0048] S11. Collect optical signal data during the welding process, where the optical signal includes visible light L1, infrared light L2 and reflected light L3. The collected optical signal data is used as a template library, where the sampling rate is 200,000 Hz and the duration of each signal is 0.2 s.

[0049] 100 template library data were collected by photoelectric sensors during laser welding, of which 5 were known to be abnormal after inspection, and their corresponding data numbers were: 10, 12, 13, 42, and 53. Taking the thermal radiation signal as an example, the data waveform collected as the initial template library is as follows: Figure 3 As shown, 3 abnormal welding explosion point data (10, 12, 13) are marked, which are manifested as local increase and bulge of the waveform, and 2 abnormal welding penetration data (42, 53) are manifested as the overall rise or fall of the waveform.

[0050] S12, performing data preprocessing on the collected welding multi-dimensional optical signals.

[0051] Furthermore, the step S12 includes:

[0052] S121. Perform equal-length interception on all three-channel data of visible light L1, infrared light L2 and reflected light L3 collected in the template library to intercept valid data in the actual welding process, and each data contains 40,600 data points.

[0053] S122, filtering and standardizing the intercepted equal-length light signal data, and stacking the three-channel light signals as a single welding sample signal x(n), where n is a data point.

[0054] S13. Continuously collect normal welding data to construct a template library D of data to be identified.

[0055] S2. Extract multi-channel local features from the data template library through a sliding time window.

[0056] Specifically, the step S2 includes:

[0057] S21. For the three-channel optical signal in the constructed template library, sequence segmentation is performed through a sliding time window; specifically including:

[0058] For the signal x(n), we select a time window size of N = 1000, starting from the beginning of the signal, and sliding S = 500 data points each time until the end of the signal. The signal of the kth window can be expressed as x k (n)=x(n+kS)·w(n), where w(n) is a window function, and a rectangular window, a Hanning window, a Hamming window, etc. can be used.

[0059] S22, performing feature extraction on each time window signal after segmentation; specifically comprising:

[0060] Feature extraction is calculated using a multi-feature weighted approach: F m (k) = Q m,1 ·mean(x k )+Q m,2 ·[max(x k )-min(x k )]+Q m,3 ·rms(x k )+Q m,4 ·std(x k )+Q m,5 ·kur(x k ), where m represents the optical signal channel, Q m,1 To Qm,5 is the weight of each feature, mean(x k )、[max(x k )-min(x k )]、rms(x k )、std(x k ) and kur(x k ) represent the signal mean, peak-to-peak value, root mean square, standard deviation and kurtosis respectively. In a specific embodiment, Q m,1 To Q m,5 They are 0.5, 0.1, 0.2, 0.1 and 0.1 respectively.

[0061] S23, repeat the above steps S21 and S22 for all sample signals in the template library, and construct a sample library F for each time window. k,k＝1,2,...K .

[0062] S3. Detect and remove abnormal data based on the isolation forest algorithm and local features.

[0063] Specifically, step S3 includes:

[0064] S31. The sample library in each time window is divided into random sample sets for multiple times to construct an isolation forest for the corresponding time window.

[0065] Furthermore, the step S31 includes:

[0066] S311, randomly selecting a certain number of samples from the sample library to form a sample set;

[0067] S312, randomly select a feature F from the sample set m (k) determining a split value and recursively splitting the data until a maximum depth is reached or only one sample is left in a node, so as to construct an isolation tree;

[0068] S313, repeat the above steps to construct multiple isolated trees to form an isolation forest.

[0069] S32, calculating the average path length of each sample on all isolated trees in the isolation forest to determine the local anomaly score; specifically including:

[0070] For each data point, calculate the average path length across all trees in the forest. The shorter the path, the more likely the data point is an anomaly. The local isolation forest algorithm calculates the local anomaly score s using the following formula: k (x) = 2 -E(h(x)) / c(x), where h(x) represents the average path length of all isolated trees in the forest, E(h(x) is the expected value of h(x), and c(n) is the average path length of the binary search tree when the search is unsuccessful, which can be calculated by the following formula: c(n) = 2H n-1 -2(n-1) / n,n>2; c(n)=1,n=2; c(n)=0,n=1; where, H n is the nth harmonic number, which can be approximated by using the natural logarithm plus Euler's constant. n ≈ln(n)+0.5772156649, local anomaly score s k The smaller (x) is, the higher the possibility that the data is abnormal in this sliding window.

[0071] S33, repeating the above steps for the sample libraries in all time windows to obtain the local anomaly score of each template library data in all time windows and merging them, thereby obtaining the overall anomaly score of all template library data; specifically including:

[0072] For K sample libraries F k,k＝1,2,...K For each of the above steps S31 and S32, the above steps are repeated until the anomaly scores of all sliding windows of all samples are calculated. For each welding data sample, the minimum anomaly score of all its sliding windows is calculated as the overall anomaly score of the welding data sample.

[0073] S34. Construct a decision function based on the overall abnormality score and the contamination rate, calculate the threshold and judge abnormal data or normal data to eliminate abnormal data; specifically including:

[0074] By constructing the decision function Dec(x)=S x -Delta, used to make decisions on the final result, where Delta is the contamination rate, which is the expected proportion of outliers in the data set. If the contamination rate is set to "auto", the algorithm automatically determines a suitable threshold; if the contamination rate is not "auto", the threshold can be calculated directly based on the contamination rate. For example, if the contamination rate is 0.1, then the threshold can be set to a value that causes 10% of the samples in the data set to be labeled as anomalies. If the output of the decision function is less than the threshold, the sample is labeled as anomaly (-1); otherwise, it is labeled as normal (1).

[0075] To facilitate comparison of results, the traditional isolation forest algorithm is used to identify anomalies in the data of this embodiment and the local isolation forest algorithm proposed in the embodiment of this application is used to compare the effects. The anomaly scores of each data calculated by the algorithm are compared with the threshold values. Figure 4As shown in the figure above, the recognition result of the local isolation forest algorithm proposed in the embodiment of the present application is shown, wherein the dots represent the abnormality score of each data, the dotted horizontal line is the set decision threshold, and the method proposed in the embodiment of the present application can recognize all 5 abnormal data (10, 12, 13, 42, 53), and the abnormality recognition accuracy rate is 100%.

[0076] The traditional isolation forest algorithm can only identify two abnormal data (42, 53), which just correspond to the data with abnormal melting depth with overall change in data size. It is unable to effectively identify the abnormal data of local change in burst point, and the abnormal identification accuracy is 40%.

[0077] By comparison, the method proposed in the embodiment of the present application can more effectively adapt to the actual working conditions of laser welding abnormality detection data to complete more accurate abnormal data identification.

[0078] For welding samples that are determined to be abnormal, they can be removed from the template library, thereby enhancing the accuracy and reliability of the template library in welding process detection.

[0079] According to a method for identifying abnormal signals for monitoring a laser welding process provided in an embodiment of the present application, it is possible to more accurately identify and eliminate noise and abnormal signals generated during the welding process due to weldment abnormalities, equipment failures, or operational errors, reduce the interference of these abnormalities on the construction of the template library, and make the template library closer to the ideal normal state. This improves the accuracy of identifying and eliminating abnormal signals, and thus enables a more accurate judgment of the welding quality in subsequent welding quality assessments, thereby ensuring that the welding process monitoring system has higher accuracy and reliability in quality detection, and ultimately ensuring that the welding process is stable and that product quality meets standards.

[0080] According to some embodiments of the present application, a laser welding process monitoring abnormal signal recognition system is also provided, which is applied to the above-mentioned method, including:

[0081] An acquisition module is used to obtain a template library of optical signal data of a laser welding process;

[0082] A feature extraction module is used to extract multi-channel local features from a data template library through a sliding time window;

[0083] The detection and elimination module is used to detect and eliminate abnormal data based on the isolation forest algorithm and local features.

[0084] In the description of the embodiments of the present application, unless otherwise clearly specified and limited, technical terms such as "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, and it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the embodiments of the present application can be understood according to the specific circumstances.

[0085] The above description is only a detailed description of the preferred embodiments and principles of the present application. For ordinary technicians in this field, there will be changes in the specific implementation methods based on the ideas provided by the present invention, and these changes should also be regarded as the scope of protection of the present application.

Claims

1. A method for identifying abnormal signals during laser welding process monitoring, characterized in that: The following steps are involved: S1, obtaining a light signal data template library for the laser welding process; S2, extracting multi-channel local features from the data template library through a sliding time window; S3. Detect and remove abnormal data based on the isolation forest algorithm and local features.

2. The method for identifying abnormal signals during laser welding process monitoring according to claim 1, characterized in that: The step S1 comprises: S11, collecting optical signals during welding, where the optical signals include visible light, infrared light and reflected light; S12, performing data preprocessing on the collected welding multi-dimensional optical signal; S13. Continuously collect normal welding data to build a template library of data to be identified.

3. The method for identifying abnormal signals during laser welding process monitoring according to claim 2, characterized in that: The step S12 comprises: S121, performing equal-length interception on all three-channel data of visible light, infrared light and reflected light collected in the template library to intercept valid data in the actual welding process; S122, filtering and standardizing the intercepted equal-length light signal data, and stacking the three-channel light signals as a single welding sample signal.

4. The method for identifying abnormal signals during laser welding process monitoring according to claim 1, characterized in that: The step S2 comprises: S21, for the optical signal in the constructed template library, sequence segmentation is performed through a sliding time window; S22, performing feature extraction on each time window signal after segmentation; S23, repeat the above steps cyclically for all sample signals in the template library, and construct a sample library for each time window.

5. The method for identifying abnormal signals during laser welding process monitoring according to claim 1, characterized in that: The step S3 comprises: S31, by splitting the sample library in each time window multiple times into random sample sets, so as to construct an isolation forest corresponding to the time window; S32, calculating the average path length of each sample on all isolated trees in the isolation forest to determine the local anomaly score; S33, repeating the above steps for the sample libraries in all time windows to obtain the local anomaly score of each template library data in all time windows and fusing them to obtain the overall anomaly score of all template library data; S34. Construct a decision function based on the overall anomaly score and the contamination rate, calculate the threshold and judge abnormal data or normal data to eliminate the abnormal data.

6. The method for identifying abnormal signals during laser welding process monitoring according to claim 5, characterized in that: The step S31 comprises: S311, randomly selecting a certain number of samples from the sample library to form a sample set; S312, randomly selecting a feature from the sample set and determining a split value to construct an isolated tree; S313, repeat the above steps to construct multiple isolated trees to form an isolation forest.

7. A laser welding process monitoring abnormal signal recognition system, characterized in that: The method as claimed in any one of claims 1 to 6 comprises: An acquisition module is used to obtain a template library of optical signal data of a laser welding process; A feature extraction module is used to extract multi-channel local features from a data template library through a sliding time window; The detection and elimination module is used to detect and eliminate abnormal data based on the isolation forest algorithm and local features.

Images