A method and system for collecting machining data of a broaching tool

Abstract

The present application relates to the technical field of tool machining data acquisition, and provides a broaching tool machining data acquisition method and system, the method comprising: acquiring original information flow from various mechanical sensors, and time-aligning the original information flow to obtain aligned original information flow; based on the aligned original information flow, identifying transient events and extracting multi-dimensional features of the transient events; constructing a multi-channel space-time correlation fingerprint library; based on the transient events and the multi-dimensional features, constructing real-time multi-channel event sequences on each channel, and matching the real-time multi-channel event sequences with the multi-channel space-time correlation fingerprint library to obtain fingerprint matching results; based on real machining event fingerprints and electromagnetic noise fingerprints, and in combination with the fingerprint matching results, distinguishing real machining events from electromagnetic noise to identify broaching tool states and machining abnormalities. The present application has the effect of improving the accuracy of broaching tool state identification and machining abnormality diagnosis.

Description

Technical Field

[0001] This invention relates to the technical field of data acquisition for tool processing, and specifically to a method and system for acquiring data for broaching tool processing. Background Technology

[0002] In the field of modern precision manufacturing, real-time and precise monitoring of the processing status of broaching tools is crucial to ensuring product quality and production efficiency.

[0003] In an automated precision broaching production line, a multi-channel information acquisition system is deployed near the high-frequency induction heating equipment. Although the signal transmission lines employ multi-layered shielding, long-term exposure to strong electromagnetic radiation and the compact wiring gradually reduce their interference resistance. Electromagnetic noise generated by the high-frequency induction heating equipment begins to infiltrate and superimpose onto the raw information stream collected by some sensitive sensors. While the amplitude of these superimposed noise signals is usually small, their frequency characteristics are similar to signals generated by microscopic events occurring during actual cutting processes, making it difficult for the central processing unit to effectively distinguish between electromagnetic noise and real machining events.

[0004] Meanwhile, the power supply stability of the distributed acquisition modules, which provide stable power to the various sensors and perform preliminary information processing, begins to be affected by the instantaneous voltage drop in the mains grid caused by the start-up and shutdown of the induction heating equipment. This may cause slight, asynchronous drifts in the timing references within some acquisition modules, gradually disrupting the originally precise time correlation between the various information streams, resulting in millisecond-level misalignments. The information stream received by the central processing unit thus exhibits two levels of degradation: first, the information content of some channels is contaminated by indistinguishable electromagnetic noise; second, the time correspondence between all channels becomes unreliable. This puts the tool condition assessment model, which relies on cross-validation of multi-channel information, in a predicament, making it unable to accurately identify the broaching tool condition and machining anomalies.

[0005] Over time, this inaccurate and contradictory diagnostic information leads operators to doubt and distrust the machining status monitoring system, potentially causing tool failure during machining, resulting in serious consequences such as workpiece scrap and production line shutdown. This highlights the urgent need for a broaching tool machining data acquisition method and system that can intelligently identify and eliminate environmental interference while maintaining accurate time alignment of multi-channel information in complex industrial environments.

[0006] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0007] This application discloses a broaching tool machining data acquisition method and system, which aims to solve the problem that in complex industrial production environments, external interference sources have a significant impact on sensitive information acquisition systems, causing contamination of the acquired raw information stream and disrupting the temporal correlation between multi-channel information, thereby affecting the accuracy of tool condition assessment and the reliability of machining anomaly identification.

[0008] The technical solution of this application is as follows:

[0009] In a first aspect, this application discloses a method for acquiring machining data from broaching tools, including:

[0010] The raw information streams from various mechanical sensors are acquired, and the raw information streams are time-aligned through local power buffering and inter-module high-frequency synchronization mechanisms to obtain aligned raw information streams.

[0011] Based on the aligned original information stream, transient events are identified, and multi-dimensional features of the transient events are extracted;

[0012] Construct a multi-channel spatiotemporal correlation fingerprint database containing fingerprints of real processing events and electromagnetic noise fingerprints;

[0013] Based on transient events and multi-dimensional features, a real-time multi-channel event sequence is constructed on each channel, and the real-time multi-channel event sequence is matched with a multi-channel spatiotemporal correlation fingerprint database to obtain fingerprint matching results;

[0014] Based on the fingerprints of real machining events and electromagnetic noise, and combined with the fingerprint matching results, real machining events are distinguished from electromagnetic noise in order to identify the broaching tool status and machining anomalies.

[0015] This technical solution effectively addresses the problem that external interference sources significantly impact sensitive information acquisition systems in complex industrial environments, leading to contamination of the acquired raw information stream and disruption of the temporal correlation between multi-channel information. This improves the accuracy of tool condition assessment and the reliability of machining anomaly identification.

[0016] Furthermore, the steps of constructing a multi-channel spatiotemporal correlation fingerprint library containing fingerprints of real processing events and electromagnetic noise also include:

[0017] Add and activate several non-contact electromagnetic field sensors, and achieve time alignment of the raw information streams of all non-contact electromagnetic field sensors.

[0018] Real-time monitoring of the raw information flow of all sensor channels, identification of transient events, and extraction of multi-dimensional features of transient events;

[0019] When a mechanical sensor detects a transient mechanical event, the information from a non-contact electromagnetic field sensor within a preset time window before the transient mechanical event occurs is simultaneously analyzed. A causal correlation analysis is then performed to obtain the causal correlation analysis results, in order to determine whether the transient mechanical event was caused by a transient change in a local electromagnetic field event.

[0020] If the causal correlation analysis results indicate that there is a causal relationship between the transient mechanical event and the local electromagnetic field event, and the local electromagnetic event occurs before the transient mechanical event, then the transient mechanical event is determined to be electromagnetically induced structural resonance interference, and the structural resonance interference is removed from the valid information.

[0021] If the causal correlation analysis results indicate that there is no causal relationship between the transient mechanical event and the local electromagnetic field event, then the transient mechanical event is compared with the fingerprint of the actual machining event to obtain the fingerprint comparison result. If the fingerprint comparison result indicates that the matching degree meets the standard, then it is determined to be tool micro-damage.

[0022] This technical solution can effectively identify and eliminate electromagnetically induced structural resonance interference, avoiding misjudgment of actual machining events. It can also more accurately identify microscopic damage to cutting tools, improving the accuracy and reliability of data acquisition.

[0023] Based on the above, this application further proposes that the steps of constructing a multi-channel spatiotemporal correlation fingerprint database containing real processing event fingerprints and electromagnetic noise fingerprints also include:

[0024] By sensing the working parameters of broaching and monitoring the structural response of the machine tool, the working parameter information and structural response information are obtained.

[0025] Based on the operating condition parameter information and structural response information, assess the deviation between the current operating condition and the standard operating condition when the fingerprint was constructed;

[0026] Based on the deviation, the fingerprints of real processing events and electromagnetic noise fingerprints are adaptively adjusted to obtain an adaptively adjusted fingerprint database.

[0027] The adaptively adjusted fingerprint database is used to match real-time multi-channel event sequences.

[0028] This technical solution enables adaptive adjustments to the fingerprint database based on actual working conditions, improving the accuracy and robustness of fingerprint matching and thus better adapting to complex and ever-changing processing environments.

[0029] Furthermore, the steps for constructing a multi-channel spatiotemporal correlation fingerprint library containing fingerprints of real processing events and electromagnetic noise also include:

[0030] Determine the degree of matching between transient events and various types of fingerprints, and mark transient events that do not match the fingerprints of real processing events and electromagnetic noise fingerprints as unknown abnormal events;

[0031] Analyze production line log information before and after the occurrence of unknown abnormal events to identify external operations or equipment status changes that are temporally related to the unknown abnormal events;

[0032] Extract the multi-channel information flow features of unknown abnormal events and cross-compare them with the identified external operations or equipment status changes to obtain the cross-comparison results;

[0033] If the cross-comparison results indicate that the multi-channel information flow characteristics of the unknown abnormal event are repeatedly correlated with the identified external operation or equipment status changes, then the operator is prompted to confirm, and the operator confirmation result is obtained.

[0034] Based on the operator's confirmation, the multi-channel information flow characteristics of the unknown abnormal event and its associated external operation or equipment status changes are added as new interference fingerprints to the multi-channel spatiotemporal correlation fingerprint database.

[0035] This technical solution enables the identification and learning of new, unknown, and abnormal events, which are then added to the fingerprint database as new interference fingerprints. This continuously improves the fingerprint database and enhances the system's ability to identify unknown anomalies.

[0036] In some preferred embodiments, the steps of identifying transient events based on the aligned original information stream and extracting multi-dimensional features of the transient events include:

[0037] Multi-scale time-frequency decomposition is performed on the aligned original information stream to obtain signal components with different frequencies and time resolutions;

[0038] Based on the signal components, identify and mark the periods in which potential transient events occur;

[0039] Based on the time period of transient events, cross-correlation analysis is performed on the signal components of different sensor channels to identify signal overlap regions.

[0040] In the signal overlap region, based on the physical propagation characteristics of signals from different sensor channels and the priority relationship between event types, aliasing events are separated to obtain each independent transient event;

[0041] For each independent transient event, extract the start time, duration, peak amplitude, energy distribution, frequency distribution, and waveform morphology of the independent transient event.

[0042] This technical solution enables refined processing of the original information stream, effectively separating aliased events and extracting rich features of independent transient events, laying the foundation for subsequent fingerprint matching and event recognition.

[0043] As an optional approach, in the signal overlap region, the steps to separate aliasing events and obtain each independent transient event based on the physical propagation characteristics of signals from different sensor channels and the priority relationship between event types include:

[0044] In the signal overlap region, based on the physical propagation characteristics of known event types, events conforming to known propagation laws are initially separated by inverting the signal propagation path;

[0045] After the initial separation is completed, residual analysis is performed on the remaining aliased signals to identify abnormal time-frequency characteristics;

[0046] The abnormal time-frequency features are compared with the recent equipment start-up or shutdown or operating condition change information recorded in the production line log to obtain the component feature association comparison results.

[0047] If the component feature correlation comparison results indicate that there is a time correlation between the abnormal time-frequency features and the recent equipment start-up / shutdown or operating condition change information, then the abnormal time-frequency features are marked as potential new interferences, and the operator is prompted to confirm, and the operator confirmation result is obtained.

[0048] Based on the operator's confirmation, the characteristics of potential new interferences and the associated external operations or equipment status changes are added as new interference fingerprints to the multi-channel spatiotemporal correlation fingerprint database, and are given priority for identification and removal in subsequent separation.

[0049] This technical solution can further refine the separation process of aliasing events, identify and learn potential new types of interference, and add them to the fingerprint database, thereby improving the system's ability to identify and eliminate new types of interference.

[0050] To enhance functionality, after the initial separation, residual analysis is performed on the remaining aliased signals to identify abnormal time-frequency characteristics. This process includes the following steps:

[0051] Multi-channel independent component analysis was performed on the remaining aliased signal to obtain several independent signal components;

[0052] Perform time-frequency analysis on each independent signal component to identify the corresponding time-frequency characteristics;

[0053] Based on time-frequency characteristics, independent signal components are clustered, and signal components with similar time-frequency characteristics are grouped into one category to form a feature cluster of potential unknown interference sources.

[0054] Feature extraction is performed on each feature cluster to obtain the abnormal time-frequency features of each independent unknown interference source.

[0055] This technical solution enables a more in-depth analysis of the remaining aliased signals, identifies potential unknown interference sources, and extracts their abnormal time-frequency characteristics, providing a more refined basis for subsequent interference identification and elimination.

[0056] To improve the scheme, the steps for extracting features from each feature cluster to obtain the anomalous time-frequency features of each independent unknown interference source include:

[0057] High-resolution time-frequency analysis is performed on the signal components within the feature cluster to obtain high-resolution time-frequency features;

[0058] Calculate the differences in spatial angle of arrival and energy attenuation path of signal components within each feature cluster on different sensors to obtain information on the differences in spatial angle of arrival and energy attenuation path.

[0059] By integrating high-resolution time-frequency features, spatial angle of arrival, and energy attenuation path differences, a multi-dimensional feature vector is constructed.

[0060] The multidimensional feature vectors are subjected to differential enhancement processing to obtain multidimensional feature vectors with enhanced differential characteristics.

[0061] Based on the multidimensional feature vectors enhanced by difference, abnormal time-frequency features are generated for each feature cluster.

[0062] This technical solution enables the extraction and fusion of features from unknown interference sources from multiple dimensions, followed by differential enhancement processing, thereby generating more distinctive abnormal time-frequency features and improving the accuracy of identifying unknown interference sources.

[0063] To optimize the structure, the multidimensional feature vectors undergo differential enhancement processing. The steps to obtain the differentially enhanced multidimensional feature vectors include:

[0064] Real-time statistical analysis is performed on the feature dimensions of each multidimensional feature vector to calculate the instantaneous rate of change and deviation of the feature dimensions;

[0065] Identify instantaneous pulse interference characteristics based on instantaneous rate of change and deviation.

[0066] Based on the characteristics of transient pulse interference, the weighting coefficients of the difference enhancement algorithm are adjusted to reduce the weight of the corresponding dimension of transient pulse interference characteristics;

[0067] By applying nonlinear amplification to dimensions other than the instantaneous pulse interference characteristics, a multidimensional feature vector with enhanced difference is obtained.

[0068] This technical solution enables differential enhancement of transient pulse interference features, reducing their impact on the overall feature vector while amplifying features in other dimensions, thereby improving the discriminability and robustness of the feature vector.

[0069] Secondly, this application also discloses a broaching tool machining data acquisition system for performing broaching tool machining data acquisition, including:

[0070] The raw information acquisition module is used to acquire raw information streams from various mechanical sensors and to perform time alignment on the raw information streams through local power buffering and high-frequency synchronization mechanism between modules to obtain aligned raw information streams.

[0071] The transient event recognition module is used to identify transient events based on the aligned raw information stream and extract multi-dimensional features of the transient events;

[0072] The fingerprint database construction module is used to build a multi-channel spatiotemporal correlation fingerprint database containing fingerprints of real processing events and electromagnetic noise fingerprints;

[0073] The fingerprint matching execution module is used to construct real-time multi-channel event sequences on each channel based on transient events and multi-dimensional features, and match the real-time multi-channel event sequences with a multi-channel spatiotemporal correlation fingerprint database to obtain fingerprint matching results.

[0074] The matching result recognition module is used to distinguish between real machining events and electromagnetic noise based on the fingerprint of real machining events and the fingerprint of electromagnetic noise, in order to identify the broaching tool status and machining abnormalities.

[0075] This technical solution provides a system for acquiring machining data from broaching tools. This system effectively addresses the problem that external interference sources significantly impact sensitive information acquisition systems in complex industrial environments, leading to contamination of the acquired raw information stream and disruption of the temporal correlation between multi-channel information. This improves the accuracy of tool condition assessment and the reliability of machining anomaly identification.

[0076] Beneficial Effects: The broaching tool machining data acquisition method disclosed in this application acquires raw information streams from various mechanical sensors and uses local power buffering and inter-module high-frequency synchronization mechanisms to time-align the raw information streams. This effectively solves the problem of unstable power supply to distributed acquisition modules in complex industrial environments, which disrupts the time correlation between various information streams, ensuring accurate time alignment of multi-channel information. Based on this, the method can identify transient events and extract multi-dimensional features from the aligned raw information streams, and construct a multi-channel spatiotemporal correlation fingerprint database containing fingerprints of real machining events and electromagnetic noise. By matching real-time multi-channel event sequences with the fingerprint database and combining real machining event fingerprints and electromagnetic noise fingerprints, it can effectively distinguish between real machining events and electromagnetic noise. This series of technical solutions overcomes the shortcomings of existing technologies in effectively distinguishing electromagnetic noise from real machining events, avoiding interference from electromagnetic noise on tool condition assessment, thereby significantly improving the accuracy and reliability of broaching tool condition identification and machining anomaly diagnosis, and effectively avoiding production accidents and economic losses caused by misjudgments. Attached Figure Description

[0077] Figure 1 This is a flowchart of a broaching tool machining data acquisition method according to one embodiment of the present invention;

[0078] Figure 2 This is a flowchart of a broaching tool machining data acquisition method according to another embodiment of the present invention;

[0079] Figure 3 This is a system block diagram of a broaching tool machining data acquisition system according to another embodiment of the present invention;

[0080] Explanation of reference numerals in the attached figures:

[0081] 1. Broaching tool machining data acquisition system; 11. Raw information acquisition module; 12. Transient event recognition module; 13. Fingerprint database construction module; 14. Fingerprint matching execution module; 15. Matching result recognition module. Detailed Implementation

[0082] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0083] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0084] In modern precision manufacturing, real-time and accurate monitoring of the broaching tool's machining status is crucial for ensuring product quality and production efficiency. However, in complex industrial production environments, external interference sources, such as high-frequency electromagnetic equipment, often significantly impact sensitive information acquisition systems, contaminating the raw information stream and disrupting the temporal correlation between multi-channel information. This decline in information quality directly affects the accuracy of tool condition assessment and the reliability of machining anomaly identification, potentially leading to production accidents and economic losses. Failure to address these issues can result in sudden broaching tool failure during machining, causing serious consequences such as workpiece scrap and production line shutdowns. This underscores the urgent need for a broaching tool machining data acquisition method and system capable of intelligently identifying and eliminating environmental interference while maintaining precise time alignment of multi-channel information in complex industrial environments.

[0085] In response, this application proposes a method for acquiring machining data using broaching tools, combined with... Figure 1 As shown, it includes:

[0086] S1: Acquire raw information streams from various mechanical sensors, and perform time alignment on the raw information streams through local power buffering and inter-module high-frequency synchronization mechanism to obtain aligned raw information streams;

[0087] S2, based on the aligned original information stream, identifies transient events and extracts multi-dimensional features of the transient events;

[0088] S3, construct a multi-channel spatiotemporal correlation fingerprint library containing fingerprints of real processing events and electromagnetic noise fingerprints;

[0089] S4. Based on transient events and multi-dimensional features, a real-time multi-channel event sequence is constructed on each channel, and the real-time multi-channel event sequence is matched with a multi-channel spatiotemporal correlation fingerprint database to obtain fingerprint matching results;

[0090] S5, based on the fingerprints of real machining events and electromagnetic noise, and combined with the fingerprint matching results, distinguishes between real machining events and electromagnetic noise in order to identify the broaching tool status and machining anomalies.

[0091] To make the technical solution of this application easier and clearer to understand, some key terms and implementation environments involved will be explained below.

[0092] "Raw information flow" refers to the unprocessed raw data directly collected by various mechanical sensors, such as vibration signals, acoustic emission signals, force signals, and temperature signals. These signals usually exist in analog or digital form and reflect the physical state during the broaching process.

[0093] "Transient events" refer to events that occur during machining processes and are short in duration but exhibit drastic changes in amplitude or frequency, such as the instant the tool contacts the workpiece, microscopic chipping, or chip breakage. These events often carry important information about the machining status.

[0094] "Multi-dimensional features" refer to multiple quantitative indicators of different dimensions extracted from transient events, such as the event's start time, duration, peak amplitude, total energy, dominant frequency, bandwidth, and waveform shape parameters (such as rise time and fall time). These features are used to comprehensively describe the characteristics of transient events.

[0095] "Real machining event fingerprint" refers to a multi-channel spatiotemporal feature pattern obtained through experiments or simulations that corresponds to specific events (such as normal cutting or the initial stage of tool wear) during the normal machining process of broaching tools.

[0096] "Electromagnetic noise fingerprint" refers to a specific multi-channel spatiotemporal characteristic pattern that appears in sensor signals caused by external electromagnetic interference sources (such as high-frequency induction heating equipment or motor startup).

[0097] The "Multi-channel Spatiotemporal Correlation Fingerprint Database" is a database that stores fingerprints of various real processing events and electromagnetic noise fingerprints. Each fingerprint contains time-series features and spatial correlation information on different sensor channels.

[0098] "Real-time multi-channel event sequence" refers to the real-time arrangement and combination of transient events and their multi-dimensional features on different sensor channels by processing the aligned raw information stream during actual processing.

[0099] The implementation environment of this application is typically a broaching production line, which is equipped with various mechanical sensors (such as accelerometers, acoustic emission sensors, force sensors, temperature sensors, etc.) as well as data acquisition and processing units.

[0100] The main features of the broaching tool machining data acquisition method of this application can be further explained.

[0101] First, regarding the acquisition of raw information streams from various mechanical sensors, the raw information streams are time-aligned through local power buffering and inter-module high-frequency synchronization mechanisms to obtain aligned raw information streams. In practical applications, raw information streams can be acquired by various types of mechanical sensors. For example, piezoelectric accelerometers can be used to measure vibration signals of cutting tools or machine tool structures; acoustic emission sensors can be used to capture high-frequency elastic wave signals during tool cutting; resistance strain gauge force sensors can be used to measure cutting forces; and thermocouples or infrared sensors can be used to monitor the temperature of the machining area. These sensors transmit their respective acquired analog signals to the local data acquisition module. To ensure the time alignment of the raw information streams, the following methods can be used: one method is to equip each local data acquisition module with an independent power buffer unit, such as using a supercapacitor or a small battery pack, to cope with instantaneous fluctuations in grid voltage and ensure the stable operation of the module's internal clock; simultaneously, a high-frequency synchronization mechanism, such as using a GPS timing module or a high-precision crystal oscillator combined with a synchronization trigger signal line, ensures that the sampling clocks of all acquisition modules are strictly synchronized. Another approach is to integrate a high-precision timestamp generator into each local data acquisition module, and then timestamp each sampling point precisely before data transmission, followed by timestamp calibration and alignment in the central processing unit.

[0102] Secondly, regarding the identification of transient events based on the aligned raw information stream and the extraction of multi-dimensional features of these events, after obtaining the time-aligned raw information stream, it is necessary to identify transient events. One approach is to use a threshold-based detection method, for example, considering a transient event as occurring when the signal amplitude exceeds a preset noise threshold; alternatively, a detection method based on signal energy changes can be used, for example, calculating the short-time energy of the signal, identifying a transient event when the energy rises sharply within a short period. After identifying the transient event, its multi-dimensional features need to be extracted. For example, the event's start time, duration, peak amplitude, total energy, dominant frequency, bandwidth, and waveform shape parameters (such as rise time and fall time) can be extracted. These features can be calculated from the signal segments of the transient event using signal processing algorithms (such as Fourier transform, wavelet transform, Hilbert transform, etc.).

[0103] Secondly, regarding the construction of a multi-channel spatiotemporal correlation fingerprint database containing fingerprints of real machining events and electromagnetic noise, a fingerprint database needs to be pre-constructed to effectively distinguish between real machining events and electromagnetic noise. One approach is to conduct broaching machining under controlled experimental conditions, collecting multi-channel sensor data under different machining states (such as normal cutting, tool wear, chipping, etc.) without electromagnetic interference, and extracting multi-dimensional features of transient events to form fingerprints of real machining events. Simultaneously, under conditions with typical electromagnetic interference sources (such as high-frequency induction heating equipment, high-power motor start-stop), the machining process is simulated, collecting multi-channel sensor data affected by electromagnetic interference, and extracting multi-dimensional features of transient events to form electromagnetic noise fingerprints. These fingerprints should not only contain features of individual channels but also spatiotemporal correlation information such as time delay, amplitude ratio, and phase difference between different channels.

[0104] Next, based on transient events and multi-dimensional features, a real-time multi-channel event sequence is constructed on each channel, and this real-time multi-channel event sequence is matched with a multi-channel spatiotemporal fingerprint database to obtain fingerprint matching results. During actual processing, the system identifies transient events in real time from the aligned raw information stream and extracts their multi-dimensional features. These real-time extracted events and their features constitute a real-time multi-channel event sequence on different sensor channels. One matching method is to use a distance-based matching algorithm, such as calculating the Euclidean or Mahalanobis distance between the feature vector of the real-time multi-channel event sequence and the feature vector of each fingerprint in the fingerprint database; a smaller distance indicates a higher matching degree. Another method is to use pattern recognition-based algorithms, such as support vector machines (SVM), neural networks, or decision trees, which are pre-trained with data from the fingerprint database, and then used to classify and match the real-time multi-channel event sequence.

[0105] Finally, regarding the distinction between real machining events and electromagnetic noise fingerprints, based on fingerprint matching results, to identify broaching tool status and machining anomalies, the final differentiation and identification are performed after obtaining the fingerprint matching results. One differentiation method is that if the matching degree between the real-time multi-channel event sequence and the real machining event fingerprint is higher than that with the electromagnetic noise fingerprint, and the matching degree reaches a preset threshold, it is determined to be a real machining event. Conversely, if the matching degree with the electromagnetic noise fingerprint is higher and reaches the threshold, it is determined to be electromagnetic noise. If the matching result shows a high match with a specific real machining event fingerprint (such as tool wear fingerprint, chipping fingerprint), the corresponding broaching tool status (such as normal, slight wear, severe wear) or machining anomaly (such as chipping, chip clogging) can be identified.

[0106] Optional, combined Figure 2As shown, the steps in S3 to construct a multi-channel spatiotemporal correlation fingerprint library containing real processing event fingerprints and electromagnetic noise fingerprints also include:

[0107] S31, add and activate several non-contact electromagnetic field sensors, and achieve time alignment of the raw information streams of all non-contact electromagnetic field sensors.

[0108] S32 monitors the raw information flow of all sensor channels in real time, identifies transient events, and extracts multi-dimensional features of transient events;

[0109] S33, When the mechanical sensor detects a transient mechanical event, the information of the non-contact electromagnetic field sensor within a preset time window before the transient mechanical event occurs is analyzed simultaneously, and a causal correlation analysis is performed to obtain the causal correlation analysis results, so as to determine whether the occurrence of the transient mechanical event is caused by the instantaneous change of the local electromagnetic field event.

[0110] S34. If the causal correlation analysis results indicate that there is a causal relationship between the transient mechanical event and the local electromagnetic field event, and the local electromagnetic event occurs before the transient mechanical event, then the transient mechanical event is determined to be electromagnetically induced structural resonance interference, and the structural resonance interference is removed from the valid information.

[0111] S35. If the causal correlation analysis results indicate that there is no causal relationship between the transient mechanical event and the local electromagnetic field event, then the transient mechanical event is compared with the fingerprint of the actual machining event to obtain the fingerprint comparison result. If the fingerprint comparison result indicates that the matching degree meets the standard, then it is determined to be tool micro-damage.

[0112] Specifically, to more accurately construct a multi-channel spatiotemporal correlation fingerprint database, this application adds and activates several non-contact electromagnetic field sensors. These sensors are configured to capture electromagnetic field changes present in the processing environment, and their raw information streams are precisely time-aligned to ensure time synchronization with the mechanical sensor data. By monitoring the raw information streams of all sensor channels (including mechanical and electromagnetic field sensors) in real time, various transient events can be identified, and multi-dimensional features of these transient events can be extracted, such as start time, duration, peak amplitude, energy distribution, frequency distribution, and waveform morphology.

[0113] When a mechanical sensor detects a transient mechanical event, the system simultaneously analyzes information from a non-contact electromagnetic field sensor within a preset time window prior to the event to determine whether it was caused by electromagnetic interference. Within this time window, a causal correlation analysis is performed between the mechanical and electromagnetic events. For example, Granger causality tests, cross-correlation analysis, or causal inference methods based on physical models can be used to determine whether the transient mechanical event was caused by a transient change in a local electromagnetic field event.

[0114] If causal correlation analysis indicates a significant causal relationship between the transient mechanical event and the local electromagnetic field event, and the local electromagnetic event occurs before the transient mechanical event, then the transient mechanical event can be identified as electromagnetically induced structural resonance interference. This interference is not a genuine machining event or tool damage, and therefore needs to be removed from the valid information to avoid contaminating the fingerprint database of genuine machining events.

[0115] Conversely, if the causal correlation analysis indicates no causal relationship between the transient mechanical event and the local electromagnetic field event, it suggests that the mechanical event was not caused by electromagnetic interference. In this case, the transient mechanical event is compared with a pre-established fingerprint of real machining events. If the comparison result shows a matching degree that meets a preset standard, the transient mechanical event can be determined to be microscopic damage to the tool, thereby achieving accurate identification of the tool's condition.

[0116] As a specific implementation, suppose that during broaching, a mechanical sensor suddenly detects a high-frequency vibration transient event. At this time, the system immediately initiates synchronous analysis of data from the non-contact electromagnetic field sensor. If, within a preset time window prior to the mechanical event, the electromagnetic field sensor detects a transient and significant electromagnetic field fluctuation, and causal correlation analysis confirms that this fluctuation is the cause of the mechanical vibration—for example, an electromagnetic pulse induced by the starting of a nearby motor or a momentary discharge of a cable causing resonance in the machine tool structure—then the high-frequency vibration transient event will be classified as electromagnetically induced structural resonance interference and removed from the valid information to be analyzed. Conversely, if no electromagnetic field fluctuation causally related to the mechanical event is detected, the mechanical event will be considered a potential real mechanical event and further compared with existing tool micro-damage fingerprints. If the comparison results show a high degree of match, it can be accurately determined that micro-damage has occurred in the tool, allowing for timely and appropriate corrective measures.

[0117] Optionally, the step of constructing a multi-channel spatiotemporal correlation fingerprint library containing fingerprints of real processing events and electromagnetic noise also includes:

[0118] By sensing the working parameters of broaching and monitoring the structural response of the machine tool, the working parameter information and structural response information are obtained.

[0119] Based on the operating condition parameter information and structural response information, assess the deviation between the current operating condition and the standard operating condition when the fingerprint was constructed;

[0120] Based on the deviation, the fingerprints of real processing events and electromagnetic noise fingerprints are adaptively adjusted to obtain an adaptively adjusted fingerprint database.

[0121] The adaptively adjusted fingerprint database is used to match real-time multi-channel event sequences.

[0122] Specifically, sensing broaching machining parameters refers to acquiring real-time operating data through various sensors integrated into the broaching machine tool or from the machine tool control system. This data may include parameters such as cutting speed, feed rate, depth of cut, coolant flow rate, spindle speed, and tool load. Monitoring machine tool structural response refers to acquiring real-time physical quantities such as vibration, deformation, and temperature of the machine tool body, fixture, or workpiece through devices such as accelerometers, strain gauges, and temperature sensors. This provides comprehensive information on operating parameters and structural response, characterizing the current machining environment.

[0123] Assessing the deviation between the current operating condition and the standard operating condition used during fingerprint construction can be understood as comparing the real-time acquired operating condition parameters and structural response information with pre-defined standard operating condition data used to build the initial fingerprint database. Specifically, this can be achieved by calculating the statistical differences of various parameters (such as mean, variance, correlation, etc.), using machine learning models for operating condition classification or regression analysis, or setting threshold ranges to determine whether the current operating condition deviates from the standard operating condition and quantify the degree of deviation. The aim is to accurately identify and quantify changes in the processing environment.

[0124] In practical applications, the fingerprints of actual machining events and electromagnetic noise are adaptively adjusted based on deviations. Specifically, this involves dynamically modifying the event features stored in the fingerprint database according to the assessed degree and direction of the deviation. For example, when an increase in cutting speed is detected, the frequency features related to tool wear in the fingerprint database can be shifted upwards, or the amplitude threshold can be modified accordingly. This adjustment can be implemented using various algorithms, such as rule-based adjustment, parametric model adjustment, or updating the fingerprint feature vector through online learning. The goal is to ensure that the fingerprint database accurately matches the event features occurring under the current actual working conditions.

[0125] Therefore, using an adaptively adjusted fingerprint database to match real-time multi-channel event sequences means applying the dynamically adjusted fingerprint database to the subsequent real-time event sequence matching process. This approach ensures the accuracy and robustness of the matching process, effectively identifying broaching tool status and machining anomalies even when operating conditions change.

[0126] In some preferred embodiments, a specific example is given below. Suppose that during broaching, the machine tool's cutting speed is adjusted from an initially set standard speed (e.g., 10 m / min) to a higher speed (e.g., 15 m / min), or the hardness of the workpiece material changes. Without the adaptive adjustment mechanism of this application, a fingerprint database built based on standard speed or standard materials may not accurately identify tool wear or chipping events under new working conditions, because the acoustic, vibrational, or electromagnetic characteristics of these events may shift in frequency or amplitude due to changes in working conditions.

[0127] However, in the scheme of this application, when the cutting speed or material hardness changes, the system obtains the latest operating condition parameter information by sensing the broaching machining operating condition parameters (e.g., obtaining cutting speed data from the machine tool controller, or detecting material hardness through sensors). Simultaneously, the machine tool structure response monitoring module continuously monitors the machine tool's vibration spectrum and temperature changes. Based on this real-time information, the system assesses the deviation between the current operating condition and the standard operating condition when the fingerprint was constructed. For example, if the cutting speed increases by 50%, the system will identify this significant deviation.

[0128] Based on this deviation, the system adaptively adjusts the fingerprints of actual machining events (e.g., vibration fingerprints of tool wear) and electromagnetic noise fingerprints. Specifically, for vibration fingerprints of tool wear, the system may adjust the characteristic frequencies stored in the fingerprint database upwards based on the known relationship between cutting speed and vibration frequency, and recalibrate the threshold of vibration amplitude. For electromagnetic noise fingerprints, if changes in operating conditions alter the characteristics of certain electromagnetic interference sources, the fingerprint database will also be updated accordingly. Finally, this adaptively adjusted fingerprint database is used to match the multi-channel event sequences acquired in real time. In this way, even with increased cutting speed or changes in material hardness, the system can still accurately identify microscopic damage to the tool or machining anomalies, avoiding misjudgments caused by changes in operating conditions, thereby ensuring the accuracy and reliability of broaching tool machining data acquisition.

[0129] Optionally, the step of constructing a multi-channel spatiotemporal correlation fingerprint library containing fingerprints of real processing events and electromagnetic noise also includes:

[0130] Determine the degree of matching between transient events and various types of fingerprints, and mark transient events that do not match the fingerprints of real processing events and electromagnetic noise fingerprints as unknown abnormal events;

[0131] Analyze production line log information before and after the occurrence of unknown abnormal events to identify external operations or equipment status changes that are temporally related to the unknown abnormal events;

[0132] Extract the multi-channel information flow features of unknown abnormal events and cross-compare them with the identified external operations or equipment status changes to obtain the cross-comparison results;

[0133] If the cross-comparison results indicate that the multi-channel information flow characteristics of the unknown abnormal event are repeatedly correlated with the identified external operation or equipment status changes, then the operator is prompted to confirm, and the operator confirmation result is obtained.

[0134] Based on the operator's confirmation, the multi-channel information flow characteristics of the unknown abnormal event and its associated external operation or equipment status changes are added as new interference fingerprints to the multi-channel spatiotemporal correlation fingerprint database.

[0135] The "judging the matching degree between transient events and various fingerprint types" refers to performing similarity calculations or pattern recognition on the multi-dimensional features of the identified transient events and all real processing event fingerprints and electromagnetic noise fingerprints stored in the multi-channel spatiotemporal correlated fingerprint database. For example, methods such as Euclidean distance, cosine similarity, or machine learning classifiers can be used to evaluate the matching degree between transient events and existing fingerprints. If the matching degree between a transient event and any known fingerprint is lower than a preset threshold, the transient event is "marked as an unknown anomalous event." Here, "unknown anomalous event" refers to an event for which there is no corresponding pattern in the current fingerprint database, which may represent a novel processing anomaly, an unknown interference source, or a system failure.

[0136] The goal of "analyzing production line log information before and after an unknown anomaly to identify external operations or equipment status changes that are temporally associated with the event" is to provide contextual information for the unknown event. Production line log information can include machine tool start-up and shutdown records, tool change records, operating parameter adjustment records, maintenance records, and external environment change records. By comparing timestamps and performing correlation analysis on these log records, it is possible to identify external operations or equipment status changes that may have led to the unknown anomaly within a specific time window before and after its occurrence. For example, if a new auxiliary device starts up within a few minutes before the unknown anomaly occurs, the start-up of that auxiliary device may be associated with the event.

[0137] "Extracting multi-channel information stream features of unknown anomalous events" refers to extracting detailed features from the original information stream for the time period marked as an unknown anomalous event across different sensor channels. These features include, but are not limited to, time-domain features (such as peak value, root mean square, waveform envelope), frequency-domain features (such as dominant frequency, bandwidth, harmonic components), and time-frequency-domain features (such as wavelet coefficients, short-time Fourier transform spectrum). These features will be used for subsequent cross-matching and fingerprint construction.

[0138] "Cross-comparison with identified external operation or equipment status changes" refers to conducting in-depth correlation analysis between the extracted multi-channel information flow characteristics of unknown abnormal events and the external operation or equipment status changes identified from the production line logs. For example, it can be analyzed whether a specific external operation (such as coolant pump startup) is always accompanied by a certain vibration frequency or acoustic characteristic (multi-channel information flow characteristics of unknown abnormal events). If a "repetitive correlation" is found between the two, that is, this specific event characteristic and the external operation or equipment status change occur simultaneously multiple times, it indicates that the unknown abnormal event may not be random noise, but a systematic disturbance caused by this external factor.

[0139] The "prompt operator confirmation" feature introduces human experience and expertise for verification. When the system identifies potential duplicate associations, it will issue an alert or prompt to the operator, providing relevant data and analysis results. The operator then uses their practical experience to determine the validity of the association. The operator's confirmation is crucial to the accuracy and reliability of the fingerprint database.

[0140] The core adaptive learning mechanism of this solution is to "add the multi-channel information flow characteristics of the unknown abnormal event and its associated external operation or equipment state change as a new interference fingerprint to the multi-channel spatiotemporal correlation fingerprint database based on the operator's confirmation." Once the operator confirms the association between the unknown abnormal event and a specific external operation or equipment state change, the system encapsulates the event's characteristics and its associated contextual information into a new interference fingerprint and incorporates it into the multi-channel spatiotemporal correlation fingerprint database. Subsequently, when similar events occur again, the system can identify and classify them as known interference, thereby avoiding misjudgment.

[0141] In some preferred embodiments, a specific example is given below. Suppose that during a broaching process, the system detects a short-duration transient vibration signal with unique frequency components using mechanical sensors. After matching with existing real machining event fingerprints and electromagnetic noise fingerprints, it is found that the vibration signal does not match any known patterns in the fingerprint database and is therefore marked as an unknown anomaly. The system then automatically analyzes the production line log information before and after the event, finding that the vibration event is always accompanied by the start-up of a new high-pressure coolant pump in the workshop. The system extracts the multi-channel information flow characteristics of this unknown anomaly (e.g., energy distribution within a specific frequency range, vibration waveform characteristics) and cross-compares them with the coolant pump start-up records, finding a high degree of repetitive correlation between the two. At this point, the system will alert the operator, informing them that an unknown vibration pattern related to the start-up of the new coolant pump has been detected. Based on experience, the operator confirms that the new coolant pump does indeed produce a unique vibration, but it has no direct negative impact on the broaching quality and should be considered a new environmental disturbance. Based on the operator's confirmation, the system adds the multi-channel information flow characteristics of this vibration mode and its correlation with the startup of the new coolant pump as a new interference fingerprint to the multi-channel spatiotemporal correlation fingerprint database. Subsequently, when the new coolant pump starts again and generates similar vibrations, the system can identify it as a known interference, rather than misjudging it as tool damage or machining abnormality. This avoids unnecessary downtime for inspection, improving production efficiency and diagnostic accuracy.

[0142] Optionally, the steps of identifying transient events and extracting multi-dimensional features of transient events based on the aligned original information stream include:

[0143] Multi-scale time-frequency decomposition is performed on the aligned original information stream to obtain signal components with different frequencies and time resolutions;

[0144] Based on the signal components, identify and mark the periods in which potential transient events occur;

[0145] Based on the time period of transient events, cross-correlation analysis is performed on the signal components of different sensor channels to identify signal overlap regions.

[0146] In the signal overlap region, based on the physical propagation characteristics of signals from different sensor channels and the priority relationship between event types, aliasing events are separated to obtain each independent transient event;

[0147] For each independent transient event, extract the start time, duration, peak amplitude, energy distribution, frequency distribution, and waveform morphology of the independent transient event.

[0148] Multi-scale time-frequency decomposition of the aligned original information stream can be understood as using techniques such as wavelet transform, short-time Fourier transform, or empirical mode decomposition to perform detailed analysis of the original information stream in the time and frequency dimensions, thereby obtaining signal components with different frequencies and time resolutions at different scales. The aim is to reveal the hidden transient features and frequency components in the signal.

[0149] Furthermore, based on the signal components, potential transient event occurrence periods are identified and marked. This can be done by setting thresholds, detecting signal abrupt changes, or using machine learning algorithms to perform pattern recognition on the signal components to determine the time intervals within the signal that may contain transient events. The aim is to initially locate the range of transient events.

[0150] Furthermore, based on the time period of transient events, cross-correlation analysis is performed on the signal components of different sensor channels to identify signal overlap regions. Specifically, this involves calculating the correlation between signals from different sensor channels to discover areas where multiple sensors simultaneously respond to a certain event. The aim is to determine the spatiotemporal range where multiple transient events may occur simultaneously or influence each other.

[0151] In areas of signal overlap, aliased events are separated into individual transient events based on the physical propagation characteristics of signals from different sensor channels and the priority relationships between event types. Specifically, the propagation speed and attenuation characteristics of signals such as sound waves, vibration waves, or electromagnetic waves in different media can be utilized, combined with pre-defined physical models of event types (e.g., tool wear, chip breakage, machine tool vibration, etc.), to decouple aliased signals. For example, vibration signals caused by tool wear typically have a specific frequency range and propagation path, while chip breakage may manifest as a high-frequency impact signal. By analyzing these physical propagation characteristics and the priority levels between event types (e.g., certain critical machining events have higher priority than background noise), aliased signals can be effectively separated into independent transient events.

[0152] Finally, for each independent transient event, the start time, duration, peak amplitude, energy distribution, frequency distribution, and waveform morphology are extracted. These multi-dimensional features are key parameters for comprehensively describing transient events. The start time is used for precise timing of the event, the duration reflects the event's persistence, the peak amplitude indicates the event's intensity, the energy and frequency distributions reveal the event's physical nature, and the waveform morphology provides the event's unique "fingerprint." The purpose is to provide rich and discriminative information for subsequent event classification and state identification.

[0153] Optionally, in the signal overlap region, the steps to separate aliasing events and obtain individual independent transient events based on the physical propagation characteristics of signals from different sensor channels and the priority relationship between event types include:

[0154] In the signal overlap region, based on the physical propagation characteristics of known event types, events conforming to known propagation laws are initially separated by inverting the signal propagation path;

[0155] After the initial separation is completed, residual analysis is performed on the remaining aliased signals to identify abnormal time-frequency characteristics;

[0156] The abnormal time-frequency features are compared with the recent equipment start-up or shutdown or operating condition change information recorded in the production line log to obtain the component feature association comparison results.

[0157] If the component feature correlation comparison results indicate that there is a time correlation between the abnormal time-frequency features and the recent equipment start-up / shutdown or operating condition change information, then the abnormal time-frequency features are marked as potential new interferences, and the operator is prompted to confirm, and the operator confirmation result is obtained.

[0158] Based on the operator's confirmation, the characteristics of potential new interferences and the associated external operations or equipment status changes are added as new interference fingerprints to the multi-channel spatiotemporal correlation fingerprint database, and are given priority for identification and removal in subsequent separation.

[0159] Specifically, in areas of signal overlap, events conforming to known propagation laws are initially separated based on the physical propagation characteristics of known event types, such as the propagation speed and attenuation patterns of sound waves, vibration waves, or electromagnetic waves in a specific medium, using signal propagation path inversion techniques. Signal propagation path inversion refers to analyzing information such as the arrival time difference and amplitude differences of signals received from multiple sensors, combined with sensor layout and medium characteristics, to infer the source location and propagation path of the signals, thereby separating known types of events from the aliased signals.

[0160] After the initial separation is completed, residual analysis is performed on the remaining aliased signals that were not initially separated. Residual analysis refers to subtracting the initially separated known event signals from the original aliased signals to obtain the remaining signals. By performing time-frequency analysis on these remaining signals, abnormal time-frequency characteristics can be identified. These characteristics may manifest as abnormal energy concentrations, frequency drifts, or irregular waveforms within specific frequencies or time periods, indicating the existence of unknown or incompletely modeled events.

[0161] The identified abnormal time-frequency characteristics are compared with recent equipment start-up / shutdown or operating condition changes recorded in the production line logs. Production line logs typically contain information such as equipment maintenance, process parameter adjustments, and auxiliary equipment operating status. By comparing the occurrence time of abnormal time-frequency characteristics with the log information, it can be determined whether the abnormal time-frequency characteristics are temporally correlated with specific external operations or equipment status changes.

[0162] If the component feature correlation comparison results indicate a temporal correlation between the abnormal time-frequency feature and recent equipment start-up / shutdown or operating condition changes, then the abnormal time-frequency feature is marked as a potential novel interference. For example, if an abnormal time-frequency feature happens to appear near the start-up or shutdown time of an auxiliary device, it is very likely that the abnormality is caused by that auxiliary device. In this case, the system will prompt the operator to confirm, obtaining the operator's confirmation result. The operator's confirmation can be based on their experience or understanding of the field situation to determine whether the abnormality is indeed a novel interference.

[0163] Based on operator confirmation, if a new type of interference is identified, its characteristics and associated external operations or equipment state changes are added as new interference fingerprints to the multi-channel spatiotemporal correlation fingerprint database. This means the system can dynamically learn and update interference patterns. Once a new interference fingerprint is added, it can be prioritized for identification and removal during subsequent event separation processes, thus preventing it from affecting the identification of genuine processing events.

[0164] In some preferred embodiments, a specific example is given below. Assume that during the broaching process, the system acquires the raw information stream through mechanical sensors such as accelerometers and acoustic emission sensors. After time alignment and preliminary transient event identification, it enters the aliasing event separation stage in the signal overlap region. First, based on the preset physical propagation characteristics of known machining events such as tool cutting and tool wear, the system uses signal propagation path inversion technology to initially separate these event signals that conform to known patterns. However, after the initial separation, the system performs residual analysis on the remaining aliased signals and discovers an abnormal time-frequency feature that continuously appears within a specific frequency range and time period. This feature does not conform to any known machining events or electromagnetic noise fingerprints.

[0165] At this point, the system automatically compares the occurrence time of this abnormal time-frequency characteristic with the production line logs. The log information shows that a new coolant pump was started shortly before the abnormal time-frequency characteristic appeared. Based on the time correlation comparison results, the system determines that the abnormal time-frequency characteristic and the start-up of the coolant pump are time-related and marks it as a potential new type of interference. Subsequently, the system will prompt the operator, requesting confirmation. Based on their on-site experience and understanding of the new coolant pump's operating status, the operator confirms that the abnormal time-frequency characteristic is indeed caused by vibration or noise generated by the new coolant pump.

[0166] Based on the operator's confirmation, the system adds the characteristics of the vibration or noise generated by the coolant pump (including its abnormal time-frequency characteristics, spatial arrival angle, and differences in energy attenuation paths) and its correlation with the external operation of "coolant pump start-up" as a new interference fingerprint to the multi-channel spatiotemporal correlation fingerprint database. In subsequent broaching data acquisition, when similar time-frequency characteristics are detected again, the system will be able to prioritize identifying and eliminate this "coolant pump vibration" interference, thereby ensuring that the identification of actual tool cutting, wear, and other machining events is not affected, improving the accuracy and reliability of data acquisition.

[0167] Optionally, after the initial separation is completed, residual analysis is performed on the remaining aliased signals to identify abnormal time-frequency characteristics. This includes the following steps:

[0168] Multi-channel independent component analysis was performed on the remaining aliased signal to obtain several independent signal components;

[0169] Perform time-frequency analysis on each independent signal component to identify the corresponding time-frequency characteristics;

[0170] Based on time-frequency characteristics, independent signal components are clustered, and signal components with similar time-frequency characteristics are grouped into one category to form a feature cluster of potential unknown interference sources.

[0171] Feature extraction is performed on each feature cluster to obtain the abnormal time-frequency features of each independent unknown interference source.

[0172] The remaining aliased signals are subjected to multi-channel independent component analysis (ICA), which aims to decompose the aliased signals received from multiple sensor channels into statistically independent signal components. ICA is a blind source separation technique that aims to discover potential independent non-Gaussian components from multivariate statistical data. This analysis can effectively separate signals from different sources in the original aliased signal, even if the sources are unknown. This yields several independent signal components, each theoretically corresponding to an independent signal source or interference source.

[0173] Furthermore, time-frequency analysis is performed on each individual signal component to reveal the energy distribution and variation patterns of these components in the time and frequency dimensions. Time-frequency analysis methods, such as Short-Time Fourier Transform (STFT), Wavelet Transform, or Hilbert-Huang Transform, can be applied to each individual signal component to identify its unique time-frequency characteristics. These characteristics may include the occurrence, duration, energy intensity, and time-varying frequency patterns of specific frequency components.

[0174] Based on this, independent signal components are clustered according to their time-frequency characteristics. Clustering analysis is an unsupervised learning method used to group data points with similar characteristics. In this context, independent signal components with similar time-frequency characteristics are grouped together, forming feature clusters of potential unknown interference sources. For example, algorithms such as K-means, DBSCAN, or hierarchical clustering can be used to group components based on similarity measures of time-frequency characteristics (such as Euclidean distance, cosine similarity, etc.). Each feature cluster represents a class of potential unknown interference sources with similar time-frequency behavior.

[0175] Finally, feature extraction is performed on each feature cluster to extract unique and discriminative anomalous time-frequency features that represent the type of interference source from each cluster. These features can be statistical aggregations of the time-frequency features of all signal components within the cluster (such as mean, variance, peak value, etc.), or they can be dimensionality reduction and representation of the cluster features using machine learning methods (such as principal component analysis (PCA) and linear discriminant analysis (LDA)). In this way, anomalous time-frequency features of each independent unknown interference source can be obtained, providing a basis for subsequent identification, classification, and processing.

[0176] Optionally, the step of extracting features from each feature cluster to obtain the anomalous time-frequency features of each independent unknown interference source includes:

[0177] High-resolution time-frequency analysis is performed on the signal components within the feature cluster to obtain high-resolution time-frequency features;

[0178] Calculate the differences in spatial angle of arrival and energy attenuation path of signal components within each feature cluster on different sensors to obtain information on the differences in spatial angle of arrival and energy attenuation path.

[0179] By integrating high-resolution time-frequency features, spatial angle of arrival, and energy attenuation path differences, a multi-dimensional feature vector is constructed.

[0180] The multidimensional feature vectors are subjected to differential enhancement processing to obtain multidimensional feature vectors with enhanced differential characteristics.

[0181] Based on the multidimensional feature vectors enhanced by difference, abnormal time-frequency features are generated for each feature cluster.

[0182] Specifically, performing high-resolution time-frequency analysis on signal components within a feature cluster refers to using advanced time-frequency analysis methods such as wavelet transform, short-time Fourier transform (STFT), or Hilbert-Huang transform (HHT) to perform refined analysis on signal components within each feature cluster with higher frequency and time resolution. This allows for the capture of weak or transient features that may be missed by traditional time-frequency analysis, with the aim of revealing a more detailed time-frequency structure of potential unknown interference sources.

[0183] The calculation of the spatial arrival angle and energy attenuation path differences of signal components within each feature cluster on different sensors refers to using the advantages of multi-sensor arrays to infer the spatial location information and propagation path characteristics of the signal source by analyzing the arrival time difference, phase difference, and signal strength attenuation of the same signal component on different sensors. Its purpose is to provide important spatial dimension information for distinguishing interference sources in different spatial locations.

[0184] In practical applications, the construction of a multi-dimensional feature vector by integrating high-resolution time-frequency features, spatial angle of arrival, and energy attenuation path difference information refers to the integration of features extracted from multiple dimensions such as time, frequency, and space to form a comprehensive feature vector. This vector can more comprehensively characterize the characteristics of potential unknown interference sources, and its purpose is to provide a rich data foundation for subsequent differential enhancement processing.

[0185] Furthermore, the differential enhancement processing of multidimensional feature vectors refers to using specific algorithms to optimize the constructed multidimensional feature vectors in order to amplify the differences between different feature clusters while suppressing similarity. The purpose is to make the features of different unknown interference sources easier to distinguish in the feature space.

[0186] Therefore, generating abnormal time-frequency features for each feature cluster based on the multidimensional feature vector after difference enhancement means that the feature vector obtained after difference enhancement can more accurately and distinctly represent the abnormal time-frequency characteristics of each potential unknown interference source. Its purpose is to provide high-quality input for subsequent interference source identification and classification.

[0187] In some preferred embodiments, it is assumed that there are two potential unknown sources of interference during the broaching process: one is a slight vibration caused by a loose part inside the machine tool, whose time-frequency characteristics may be highly similar to normal machining vibrations at low resolution; the other is electromagnetic pulse interference occasionally generated by a nearby device, whose time-frequency characteristics may be superimposed with some mechanical transient events.

[0188] According to the scheme of this application, firstly, a high-resolution time-frequency analysis is performed on these aliased signals, such as by using continuous wavelet transform, which can reveal a finer harmonic structure in the vibration of loose parts, as well as the extremely short rise time and attenuation characteristics in electromagnetic pulse interference, which are difficult to capture by traditional Fourier transform.

[0189] Secondly, by deploying multiple acceleration sensors and electromagnetic field sensors at different locations on the machine tool, the differences in the spatial arrival angle and energy attenuation path of these signal components on different sensors are calculated. For example, the vibration signal of a loose part may exhibit higher energy and a faster arrival time on the sensor closer to the part, while electromagnetic pulses may exhibit a stronger response on the electromagnetic field sensor, and their propagation path differs significantly from that of mechanical vibration signals.

[0190] Next, these high-resolution time-frequency features, spatial arrival angles, and energy attenuation path differences are fused to construct a feature vector containing multi-dimensional information such as time, frequency, and space.

[0191] Finally, this multidimensional feature vector undergoes differential enhancement processing. For example, methods such as Principal Component Analysis (PCA) or Linear Discriminant Analysis (LDA), combined with specific weighting strategies, can be used to amplify the distance between different interference sources in the feature space. Through this enhancement processing, even if two interference sources are similar in some dimensions, they can be clearly distinguished in the fused multidimensional feature space, thereby generating highly discriminative anomalous time-frequency features for each feature cluster, ultimately achieving accurate identification and classification of vibrations of loose components and electromagnetic pulse interference.

[0192] Optionally, the steps to perform differential enhancement processing on the multidimensional feature vectors to obtain the differentially enhanced multidimensional feature vectors include:

[0193] Real-time statistical analysis is performed on the feature dimensions of each multidimensional feature vector to calculate the instantaneous rate of change and deviation of the feature dimensions;

[0194] Identify instantaneous pulse interference characteristics based on instantaneous rate of change and deviation.

[0195] Based on the characteristics of transient pulse interference, the weighting coefficients of the difference enhancement algorithm are adjusted to reduce the weight of the corresponding dimension of transient pulse interference characteristics;

[0196] By applying nonlinear amplification to dimensions other than the instantaneous pulse interference characteristics, a multidimensional feature vector with enhanced difference is obtained.

[0197] Specifically, the feature dimensions of a multidimensional feature vector can include, but are not limited to, time-frequency features, spatial arrival angle information, and energy attenuation path difference information. Real-time statistical analysis of these feature dimensions refers to continuously monitoring and calculating the values ​​of each dimension as the data stream continues to input, in order to obtain its dynamic characteristics. The instantaneous rate of change can be understood as the magnitude of change of the feature value relative to the previous moment or average value within a very short time window, which can be obtained, for example, by calculating the first or second derivative. The deviation indicates the degree of deviation between the current feature value and the historical average value or preset benchmark value of that dimension.

[0198] Furthermore, based on the calculated instantaneous rate of change and deviation, transient impulse interference characteristics can be identified. Transient impulse interference typically manifests as large, high-frequency changes in feature values ​​within a very short period, with deviations significantly exceeding the normal range. For example, thresholds can be set; when both the instantaneous rate of change and deviation of a certain feature dimension exceed a preset threshold, that dimension is determined to be subject to transient impulse interference. These thresholds can be trained and optimized through historical data analysis, expert experience, or machine learning methods.

[0199] Once transient impulse interference features are identified, the weighting coefficients of the difference enhancement algorithm are adjusted. Specifically, the weights of the dimensions corresponding to the transient impulse interference features are reduced. This means that in subsequent difference enhancement processing, the contribution of these interfered dimensions is weakened to reduce their negative impact on the overall feature vector. For example, an adaptive weighting strategy can be used to dynamically adjust the weights according to the intensity of the transient impulse interference; the stronger the interference, the greater the reduction in weight.

[0200] Simultaneously, to ensure that authentic and effective feature information is sufficiently enhanced, nonlinear amplification is applied to dimensions other than the instantaneous impulse interference features. Nonlinear amplification can employ various functional forms, such as the sigmoid function, ReLU function, or exponential function, aiming to significantly improve the discriminative power of non-interference dimensions without introducing excessive noise. In this way, features representing genuine processing events or potential anomalies become more prominent in the enhanced multidimensional feature vector, thereby improving the accuracy of subsequent matching and recognition.

[0201] This application also discloses a broaching tool machining data acquisition system, used to perform broaching tool machining data acquisition, combined with... Figure 3 As shown, the broaching tool machining data acquisition system 1 includes:

[0202] The raw information acquisition module 11 is used to acquire raw information streams from various mechanical sensors and to perform time alignment on the raw information streams through local power buffering and inter-module high-frequency synchronization mechanism to obtain aligned raw information streams.

[0203] The transient event recognition module 12 is used to identify transient events based on the aligned original information stream and extract multi-dimensional features of the transient events;

[0204] Fingerprint database construction module 13 is used to construct a multi-channel spatiotemporal correlation fingerprint database containing real processing event fingerprints and electromagnetic noise fingerprints;

[0205] The fingerprint matching execution module 14 is used to construct a real-time multi-channel event sequence on each channel based on transient events and multi-dimensional features, and match the real-time multi-channel event sequence with the multi-channel spatiotemporal correlation fingerprint database to obtain the fingerprint matching result;

[0206] The matching result recognition module 15 is used to distinguish between real machining events and electromagnetic noise based on the fingerprint of real machining events and the fingerprint of electromagnetic noise, and to identify the broaching tool status and machining abnormalities.

[0207] The main modules of the broaching tool machining data acquisition system of this application can be further described in detail.

[0208] First, regarding the raw information acquisition module. This module acquires raw information streams from various mechanical sensors and performs time alignment on the raw information streams through local power buffering and a high-frequency synchronization mechanism between modules to obtain an aligned raw information stream. The acquisition method of the raw information stream and the specific mechanism of time alignment have been described in the above embodiments and will not be repeated here. It should be emphasized that the raw information acquisition module can be implemented in two ways: one is by consisting of multiple independent distributed data acquisition units, each unit integrating an analog-to-digital converter, a local power buffer circuit (e.g., containing a supercapacitor or a small backup battery), and a high-precision clock synchronization circuit (e.g., based on GPS timing or the PTP protocol). Another way is that the raw information acquisition module can be a centralized high-speed data acquisition card, which connects to various mechanical sensors through multiple parallel input interfaces and achieves synchronous sampling and time alignment of all channels through an internal hardware clock and an external synchronization signal input port.

[0209] Secondly, regarding the transient event identification module. This module is used to identify transient events based on the aligned raw information stream and extract multi-dimensional features of the transient events. The identification method for transient events and the extraction method for multi-dimensional features have been described in the above embodiments, and will not be repeated here. It should be emphasized that the transient event identification module can be implemented in two ways: one is as a dedicated digital signal processor (DSP) or field-programmable gate array (FPGA) with built-in signal processing algorithms (such as threshold detection, energy change detection, wavelet transform, etc.) capable of high-speed processing of real-time data streams. The other way is that the transient event identification module can be a software program running on a general-purpose computer platform, which calls the corresponding signal processing library functions to perform offline or near-real-time analysis of the aligned raw information stream received from the raw information acquisition module.

[0210] Secondly, regarding the fingerprint database construction module. This module is used to construct a multi-channel spatiotemporal correlation fingerprint database containing fingerprints of real processing events and electromagnetic noise fingerprints. The method for constructing the fingerprint database has already been described in the above embodiments and will not be repeated here. It should be emphasized that the fingerprint database construction module can be implemented in two ways: one is as an independent database server that stores various fingerprint data obtained in advance through experiments or simulations and provides a data interface for other modules to access. Another way is that the fingerprint database construction module can be a storage unit (e.g., a solid-state drive or non-volatile memory) integrated into the main control unit, which pre-stores fingerprint data and allows the fingerprint matching execution module to directly read it through a file system or memory mapping method.

[0211] Next, regarding the fingerprint matching execution module. This module is used to construct real-time multi-channel event sequences on each channel based on transient events and multi-dimensional features, and then match these real-time multi-channel event sequences with a multi-channel spatiotemporal fingerprint database to obtain fingerprint matching results. The construction method of the real-time multi-channel event sequences and the matching algorithm with the fingerprint database have already been described in the above embodiments, and will not be repeated here. It is important to emphasize that the fingerprint matching execution module can be implemented in two ways: one is as a high-performance computing unit, such as an industrial PC or embedded controller, running a pattern recognition algorithm (such as SVM, neural network, or distance matching algorithm), capable of real-time processing and matching of multi-dimensional features of transient events. Another way is that the fingerprint matching execution module can be a distributed processing unit based on a cloud computing or edge computing architecture, where the matching algorithm is deployed on a cloud or edge server, receiving real-time event sequences via a network and returning matching results.

[0212] Finally, regarding the matching result recognition module. This module is used to distinguish between real machining events and electromagnetic noise based on the fingerprints of real machining events and electromagnetic noise, combined with the fingerprint matching results, in order to identify the broaching tool status and machining anomalies. The methods for distinguishing between real machining events and electromagnetic noise, as well as the logic for identifying the broaching tool status and machining anomalies, have already been described in the above embodiments, and will not be repeated here. It should be emphasized that the matching result recognition module can be implemented in two ways: one is as a decision support system, which applies preset decision rules or machine learning models to perform the final classification and judgment based on the matching results output by the fingerprint matching execution module. Another way is that the matching result recognition module can be a software component integrated into the human-machine interface (HMI), which visualizes the matching results to the operator and learns and optimizes based on the operator's feedback to improve the accuracy of recognition.

[0213] The above are merely embodiments of this application and are not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for acquiring machining data using broaching tools, characterized in that, include: The raw information streams from various mechanical sensors are acquired, and the raw information streams are time-aligned through local power buffering and inter-module high-frequency synchronization mechanisms to obtain aligned raw information streams. Based on the aligned original information stream, transient events are identified, and multi-dimensional features of the transient events are extracted. Construct a multi-channel spatiotemporal correlation fingerprint database containing fingerprints of real processing events and electromagnetic noise fingerprints; Based on the transient events and the multi-dimensional features, a real-time multi-channel event sequence is constructed on each channel, and the real-time multi-channel event sequence is matched with the multi-channel spatiotemporal correlation fingerprint database to obtain the fingerprint matching result; Based on the fingerprints of real machining events and electromagnetic noise, and combined with the fingerprint matching results, real machining events and electromagnetic noise are distinguished to identify broaching tool status and machining anomalies. The step of constructing a multi-channel spatiotemporal correlation fingerprint library containing real processing event fingerprints and electromagnetic noise fingerprints also includes: Add and activate several non-contact electromagnetic field sensors, and achieve time alignment of the raw information streams of all non-contact electromagnetic field sensors. Real-time monitoring of the raw information flow of all sensor channels, identification of transient events, and extraction of multi-dimensional features of the transient events; When a mechanical sensor detects a transient mechanical event, the information from a non-contact electromagnetic field sensor within a preset time window before the transient mechanical event occurs is simultaneously analyzed. A causal correlation analysis is performed to obtain the causal correlation analysis results, so as to determine whether the occurrence of the transient mechanical event is caused by the instantaneous change of a local electromagnetic field event. If the causal correlation analysis results indicate that there is a causal relationship between the transient mechanical event and the local electromagnetic field event, and the local electromagnetic event occurs before the transient mechanical event, then the transient mechanical event is determined to be electromagnetically induced structural resonance interference, and the structural resonance interference is removed from the valid information. If the causal correlation analysis results indicate that there is no causal relationship between the transient mechanical event and the local electromagnetic field event, then the transient mechanical event is compared with the fingerprint of the actual machining event to obtain the fingerprint comparison result. If the fingerprint comparison result indicates that the matching degree meets the standard, then it is determined to be tool micro-damage.

2. The broaching tool machining data acquisition method according to claim 1, characterized in that, The step of constructing a multi-channel spatiotemporal correlation fingerprint library containing real processing event fingerprints and electromagnetic noise fingerprints also includes: By sensing the working parameters of broaching and monitoring the structural response of the machine tool, the working parameter information and structural response information are obtained. Based on the operating condition parameter information and the structural response information, evaluate the deviation between the current operating condition and the standard operating condition during fingerprint construction; Based on the deviation, the fingerprints of the actual processing events and the electromagnetic noise fingerprints are adaptively adjusted to obtain an adaptively adjusted fingerprint database. The adaptively adjusted fingerprint database is used to match real-time multi-channel event sequences.

3. The broaching tool machining data acquisition method according to claim 1, characterized in that, The step of constructing a multi-channel spatiotemporal correlation fingerprint library containing real processing event fingerprints and electromagnetic noise fingerprints also includes: Determine the degree of matching between the transient event and various fingerprints, and mark transient events that do not match the real processing event fingerprint and the electromagnetic noise fingerprint as unknown abnormal events; Analyze the production line log information before and after the occurrence of the unknown abnormal event to identify the external operations or equipment status changes that are temporally associated with the unknown abnormal event; Extract the multi-channel information stream features of the unknown abnormal event and cross-compare them with the identified external operations or device state changes to obtain the cross-comparison results. If the cross-comparison result indicates that the multi-channel information flow characteristics of the unknown abnormal event are repeatedly correlated with the identified external operation or equipment status change, then the operator is prompted to confirm, and the operator confirmation result is obtained. Based on the operator's confirmation, the multi-channel information flow characteristics of the unknown abnormal event and its associated external operation or device state changes are added as new interference fingerprints to the multi-channel spatiotemporal correlation fingerprint database.

4. The broaching tool machining data acquisition method according to claim 1, characterized in that, The steps of identifying transient events based on the aligned original information stream and extracting multi-dimensional features of the transient events include: Multi-scale time-frequency decomposition is performed on the aligned original information stream to obtain signal components with different frequencies and time resolutions; Based on the signal components, identify and mark the periods in which potential transient events occur; Based on the time period of the transient event, cross-correlation analysis is performed on the signal components of different sensor channels to identify signal overlap regions; In the signal overlap region, based on the physical propagation characteristics of signals from different sensor channels and the priority relationship between event types, aliasing events are separated to obtain each independent transient event; For each independent transient event, extract the start time, duration, peak amplitude, energy distribution, frequency distribution, and waveform morphology of the independent transient event.

5. The broaching tool machining data acquisition method according to claim 4, characterized in that, The step of separating aliasing events and obtaining individual independent transient events in the signal overlap region according to the physical propagation characteristics of signals from different sensor channels and the priority relationship between event types includes: In the signal overlap region, based on the physical propagation characteristics of known event types, events conforming to known propagation laws are initially separated by inverting the signal propagation path. After the initial separation is completed, residual analysis is performed on the remaining aliased signals to identify abnormal time-frequency characteristics; The abnormal time-frequency features are compared with the recent equipment start-up or shutdown or operating condition change information recorded in the production line log to obtain the component feature association comparison results. If the component feature correlation comparison result indicates that there is a time correlation between the abnormal time-frequency feature and the recent equipment start-up / shutdown or operating condition change information, then the abnormal time-frequency feature is marked as a potential new type of interference, and the operator is prompted to confirm, and the operator confirmation result is obtained. Based on the operator's confirmation, the characteristics of the potential novel interference and the associated external operation or equipment state changes are added as new interference fingerprints to the multi-channel spatiotemporal correlation fingerprint database, and are preferentially identified and removed in subsequent separation.

6. The broaching tool machining data acquisition method according to claim 5, characterized in that, After the initial separation is completed, the remaining aliased signals are subjected to residual analysis to identify abnormal time-frequency characteristics. The steps include: Multi-channel independent component analysis was performed on the remaining aliased signal to obtain several independent signal components; Perform time-frequency analysis on each independent signal component to identify the corresponding time-frequency characteristics; Based on the time-frequency characteristics, independent signal components are clustered, and signal components with similar time-frequency characteristics are grouped into one category to form a feature cluster of potential unknown interference sources. Feature extraction is performed on each of the feature clusters to obtain the abnormal time-frequency features of each independent unknown interference source.

7. The broaching tool machining data acquisition method according to claim 6, characterized in that, The step of extracting features from each of the feature clusters to obtain the abnormal time-frequency features of each independent unknown interference source includes: A preset high-resolution time-frequency analysis is performed on the signal components within the feature cluster to obtain high-resolution time-frequency features; Calculate the differences in spatial angle of arrival and energy attenuation path of signal components within each feature cluster on different sensors to obtain information on the differences in spatial angle of arrival and energy attenuation path. By integrating the high-resolution time-frequency features, the spatial angle of arrival, and the energy attenuation path difference information, a multi-dimensional feature vector is constructed. The multidimensional feature vector is subjected to difference enhancement processing to obtain a difference-enhanced multidimensional feature vector; Based on the multidimensional feature vectors enhanced by difference, abnormal time-frequency features are generated for each feature cluster.

8. The broaching tool machining data acquisition method according to claim 7, characterized in that, The step of performing difference enhancement processing on the multidimensional feature vector to obtain the difference-enhanced multidimensional feature vector includes: Real-time statistical analysis is performed on the feature dimensions of each of the multidimensional feature vectors to calculate the instantaneous rate of change and deviation of the feature dimensions; Based on the instantaneous rate of change and the deviation, instantaneous pulse interference characteristics are identified; Based on the instantaneous pulse interference characteristics, the weighting coefficients of the difference enhancement algorithm are adjusted to reduce the weight of the dimension corresponding to the instantaneous pulse interference characteristics; By applying nonlinear amplification to dimensions other than the instantaneous pulse interference characteristics, a multidimensional feature vector with enhanced difference is obtained.

9. A broaching tool machining data acquisition system, used for performing broaching tool machining data acquisition, characterized in that, include: The raw information acquisition module is used to acquire raw information streams from various mechanical sensors, and to perform time alignment on the raw information streams through local power buffering and inter-module high-frequency synchronization mechanism to obtain aligned raw information streams. The transient event identification module is used to identify transient events based on the aligned original information stream and extract the multi-dimensional features of the transient events; The fingerprint database construction module is used to build a multi-channel spatiotemporal correlation fingerprint database containing fingerprints of real processing events and electromagnetic noise fingerprints; The fingerprint matching execution module is used to construct a real-time multi-channel event sequence on each channel based on the transient event and the multi-dimensional features, and match the real-time multi-channel event sequence with the multi-channel spatiotemporal correlation fingerprint database to obtain the fingerprint matching result; The matching result recognition module is used to distinguish between real machining events and electromagnetic noise based on the fingerprint of real machining events and the fingerprint of electromagnetic noise, in order to identify the broaching tool status and machining abnormalities. The construction of the multi-channel spatiotemporal correlation fingerprint library, which includes fingerprints of real processing events and electromagnetic noise fingerprints, includes: Add and activate several non-contact electromagnetic field sensors, and achieve time alignment of the raw information streams of all non-contact electromagnetic field sensors. Real-time monitoring of the raw information flow of all sensor channels, identification of transient events, and extraction of multi-dimensional features of the transient events; When a mechanical sensor detects a transient mechanical event, the information from a non-contact electromagnetic field sensor within a preset time window before the transient mechanical event occurs is simultaneously analyzed. A causal correlation analysis is performed to obtain the causal correlation analysis results, so as to determine whether the occurrence of the transient mechanical event is caused by the instantaneous change of a local electromagnetic field event. If the causal correlation analysis results indicate that there is a causal relationship between the transient mechanical event and the local electromagnetic field event, and the local electromagnetic event occurs before the transient mechanical event, then the transient mechanical event is determined to be electromagnetically induced structural resonance interference, and the structural resonance interference is removed from the valid information. If the causal correlation analysis results indicate that there is no causal relationship between the transient mechanical event and the local electromagnetic field event, then the transient mechanical event is compared with the fingerprint of the actual machining event to obtain the fingerprint comparison result. If the fingerprint comparison result indicates that the matching degree meets the standard, then it is determined to be tool micro-damage.

Images