A method and system for processing operation data of a laser device in an industrial internet

Abstract

The application relates to the technical field of industrial internet, and discloses a laser equipment operation data processing method and system in the industrial internet. The method comprises the following steps: collecting operation information of laser equipment, processing material attributes, cutting area optical characteristics and cutting area acoustic characteristics, and processing the collected information to obtain a historical process characteristic vector, so that multi-source heterogeneous data are integrated, and the complex state of a laser cutting process is comprehensively reflected. Based on the historical process characteristic vector, a multi-dimensional process state space is constructed, and a defect-free operation area and a known defect area are defined in the multi-dimensional process state space, so that the system can intuitively understand and quantify the risk under different operation states.

Description

Technical Field

[0001] This invention relates to the field of industrial internet technology, and in particular to a method and system for processing laser equipment operation data in the industrial internet. Background Technology

[0002] In modern industrial production, especially in fields with extremely high product quality requirements such as aerospace and medical equipment, laser cutting technology plays a crucial role. To ensure that the dimensional accuracy and surface finish of these high-value components meet stringent standards, factories commonly deploy advanced industrial internet systems to monitor and control the operation of laser equipment in real time. However, even under the monitoring of these advanced systems, when all equipment parameters appear stable and within normal ranges, some subtle, intermittent microscopic quality defects may still occur. These defects are often not caused by a single parameter anomaly, but rather by the complex and nonlinear interactions between the internal operating state of the equipment, the microscopic properties of the raw materials, and physical phenomena such as the molten pool and plasma during the cutting process. Traditional data processing methods based on thresholds or simple correlations are insufficient to effectively identify and predict these hidden early defect precursors, leading to problems being discovered only in subsequent inspection stages, resulting in high rework or scrap costs.

[0003] To delve deeper into the root causes of these elusive quality issues, the engineering team deduced that the problem might not be limited to parameters directly controlled by the machines. They believed that the microscopic properties of the material itself or deeper physical phenomena during the cutting process might not be fully reflected in the existing data. Therefore, they decided to expand the scope of data collection. First, they integrated the material traceability system with an industrial internet platform, assigning a unique identifier to each batch of raw metal sheets and recording detailed information, including the supplier, minute fluctuations in specific alloy composition (even within standard tolerances), surface treatment process details, and even internal stress data measured during incoming inspection. Second, they installed high-resolution optical sensors near the working area of ​​each laser cutting machine to capture real-time image data of the molten pool dynamics and plasma plume, while also deploying highly sensitive acoustic sensors to pick up subtle changes in sound characteristics generated during the cutting process. Their idea was that minute differences in material properties, or subtle changes in the physical processes of cutting, might be the underlying causes of these hidden defects, information that was not directly reflected by the machine's internal sensors.

[0004] In attempting to correlate newly introduced data with previously discovered subtle defects, the technical team found that simple direct correlation analyses were often ineffective. Instead, they observed a more complex phenomenon: certain types of microcracks only appeared when the material being processed came from a specific supplier and simultaneously, acoustic sensors detected a specific high-frequency resonance mode, even when machine parameters such as laser power and cutting speed appeared "normal." Similarly, localized surface roughness issues more often occurred when the purity of the cutting gas was within a specific range and a specific oscillation frequency of the molten pool was present, without a direct linear relationship to the focal location reported by the machine. These were not isolated events, but complex, context-dependent anomalies where a "normal" range for one parameter only became potentially hazardous when combined with a specific state of one or more seemingly unrelated parameters or material properties. Existing data processing methods, primarily relying on setting thresholds for individual parameters or performing simple linear regression analysis, are completely ineffective against such complex patterns involving multiple information sources, overlapping factors, and nonlinear relationships. They cannot effectively identify and quantify these "conditional" defect characteristics that indicate early failure risk.

[0005] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0006] This invention provides a method and system for processing laser equipment operation data in the Industrial Internet, which solves the technical problems of existing technologies that make it difficult to effectively identify and predict hidden early defects in the laser cutting process, and that traditional data processing methods cannot cope with complex patterns involving multiple information sources, multiple factors, and nonlinear relationships, thus leading to high rework or scrap costs.

[0007] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a method for processing laser equipment operation data in the Industrial Internet, comprising: Collect and process the operating information of the laser equipment, the properties of the processed material, the optical characteristics of the cutting area, and the acoustic characteristics of the cutting area to obtain a historical process feature vector. Based on the historical process feature vector, a multidimensional process state space is constructed, and a defect-free operation region and a known defect region are defined in the multidimensional process state space. The current process feature vector is acquired in real time as the current process state point, and the distance from the current process state point to the boundary of the known defect area is calculated to obtain the health margin. A warning boundary is set, and a risk warning is triggered when the health margin continuously reaches the warning boundary; Based on the risk direction indicated by the risk warning, parameter adjustment suggestions are generated and issued.

[0008] Preferably, the step of acquiring the current process feature vector in real time as the current process state point and calculating the distance from the current process state point to the boundary of the known defect region to obtain the health margin includes: Periodically modulate the laser pulse frequency or the auxiliary gas flow rate; Collect cutting head vibration data, microphone audio data, and molten pool image sequences; Spectral analysis is performed on the vibration data of the cutting head to obtain the energy peak value near the excitation frequency or its harmonics or sub-harmonics. Spectral analysis is performed on the microphone audio data to obtain the energy at the perturbation frequency and its subharmonics or intermodulation frequencies; Inter-frame difference and local pixel brightness fluctuation spectrum analysis are performed on the molten pool image sequence to obtain the frequency and intensity of the periodic blur pattern and the energy of a specific low-frequency band; The acquired energy peak value, energy, frequency, and amplitude are used as new feature values ​​and added to the current process feature vector to form an extended process feature vector; Based on the extended process feature vector, the distance from the current process state point to the boundary of the known defect region is recalculated to obtain the updated health margin.

[0009] Preferably, the acquisition of the laser equipment's operating information, processed material properties, optical characteristics of the cutting area, and acoustic characteristics of the cutting area to obtain a historical process feature vector includes: Construct a high-precision time synchronization network to synchronize the clocks of data acquisition devices and computing units; Collect laser equipment operation information, processed material properties, optical characteristics of the cutting area, and acoustic characteristics of the cutting area, and add timestamps; A microsecond-level time window is set, and the laser equipment operation information, the properties of the processed material, the optical characteristics of the cutting area, and the acoustic characteristics of the cutting area are aggregated. The aggregated data is then combined to form the historical process feature vector. Monitor data source timestamp deviation and dynamically adjust timestamp correction coefficients; The correction coefficients are applied to perform time alignment on the features in the historical process feature vector.

[0010] Preferably, the monitoring of data source timestamp deviation and the dynamic adjustment of the timestamp correction coefficient include: Collect the local timestamps of each data source and the timestamp of the master clock, and calculate the instantaneous deviation; When the instantaneous deviation exceeds a preset threshold, the compensation module is activated; A nonlinear compensation function is generated based on the ambient temperature fluctuation curve, equipment operating time records, and electromagnetic interference intensity. The instantaneous deviation is corrected by applying the nonlinear compensation function to obtain the corrected deviation; Perform trend analysis on the corrected deviation; The correction coefficients are adjusted based on the results of the trend analysis.

[0011] Preferably, triggering a risk warning when the health margin continuously reaches the warning boundary includes: Obtain current operating condition information; Obtain the frequency and duration at which the health margin reaches the warning boundary; Based on the current operating condition information and the frequency and duration of the health margin reaching the warning boundary, adjust the threshold of the warning boundary; Based on the adjusted warning boundary thresholds, assess the severity and urgency of the risk and trigger tiered warnings; Based on the aforementioned tiered early warning system, early warning information is pushed out. Based on the warning information and subsequent quality inspection results, the warning boundary adjustment strategy is optimized.

[0012] Preferably, the step of generating parameter adjustment suggestions based on the risk direction indicated by the risk warning and issuing the parameter adjustment suggestions includes: Based on the risk direction indicated by the risk warning, identify the dominant coupled physical phenomena in the current cutting process; Based on the dominant coupling physical phenomena, determine the set of affected device parameters and their interaction relationships; Set multi-parameter adjustment targets, including defect avoidance priority and production efficiency constraints; Within the range of the device parameter set, parameter adjustment candidates are generated based on the interaction relationship and the multi-parameter adjustment target; By simulating the impact of the parameter adjustment candidates on the dominant coupled physical phenomenon, the overall effect of the parameter adjustment candidates on defect avoidance and production efficiency is evaluated. Select the parameter adjustment candidate with the best overall effect as the parameter adjustment suggestion, and issue the parameter adjustment suggestion.

[0013] Preferably, identifying the dominant coupled physical phenomenon in the current cutting process based on the risk direction indicated by the risk warning includes: Multi-band, multi-amplitude scanning modulation of laser pulse frequency or auxiliary gas flow rate; Collect cutting head vibration data, microphone audio data, and molten pool image sequences; Nonlinear spectral response analysis is performed on the vibration data of the cutting head, the audio data of the microphone, and the image sequence of the molten pool to identify energy peaks or mode changes that occur at the harmonics, subharmonics, or intermodulation frequencies of the modulation frequency. High-order statistical feature analysis is performed on the molten pool image sequence to identify non-periodic blur or texture anomalies; Identified energy peaks, pattern changes, blurring, or texture anomalies are used as novel features; The novel features are correlated with the current process feature vector; Identify novel coupled physical phenomena that differ significantly from the responses of known normal processes; Record the characteristic combinations of the novel coupled physical phenomena and their corresponding operating condition information; The aforementioned novel coupled physical phenomenon is incorporated into the risk early warning mechanism as a new risk model.

[0014] Preferably, identifying the dominant coupled physical phenomenon in the current cutting process based on the risk direction indicated by the risk warning includes: Multi-band, multi-amplitude scanning modulation of laser pulse frequency or auxiliary gas flow rate; Collect cutting head vibration data, microphone audio data, and molten pool image sequences; Nonlinear spectral response analysis is performed on the vibration data of the cutting head, the audio data of the microphone, and the image sequence of the molten pool to identify energy peaks or mode changes that occur at the harmonics, subharmonics, or intermodulation frequencies of the modulation frequency. High-order statistical feature analysis is performed on the molten pool image sequence to identify non-periodic blur or texture anomalies; Identified energy peaks, pattern changes, blurring, or texture anomalies are used as novel features; The novel features are correlated with the current process feature vector; Identify novel coupled physical phenomena that differ significantly from the responses of known normal processes; Record the characteristic combinations of the novel coupled physical phenomena and their corresponding operating condition information; The aforementioned novel coupled physical phenomenon is incorporated into the risk early warning mechanism as a new risk model.

[0015] Preferably, the step of generating parameter adjustment suggestions based on the risk direction indicated by the risk warning and issuing the parameter adjustment suggestions includes: Based on the risk direction indicated by the risk warning, identify the dominant coupled physical phenomena in the current cutting process; Based on the dominant coupling physical phenomena, determine the set of affected device parameters and their interaction relationships; Obtain the type of material being processed, the expected product service environment, and the design life requirements, and obtain the long-term performance impact factors corresponding to the equipment parameters; Set multi-parameter adjustment targets, which include defect avoidance priority, production efficiency constraints, and long-term performance impact weights; Based on the interaction relationship, the multi-parameter adjustment target, and the long-term performance impact factor, a set of parameter adjustment candidates is generated. The effects of the parameter adjustment candidates on the dominant coupled physical phenomenon are simulated, and the comprehensive effects of the parameter adjustment candidates on defect avoidance, production efficiency and long-term product performance are evaluated in combination with the long-term performance impact factor. Select the parameter adjustment candidate with the best overall effect as the parameter adjustment suggestion, and issue the parameter adjustment suggestion.

[0016] Secondly, the present invention provides a laser equipment operation data processing system in the Industrial Internet, comprising: The acquisition end is used to collect and process the operating information of the laser equipment, the properties of the processed material, the optical characteristics of the cutting area, and the acoustic characteristics of the cutting area to obtain the historical process feature vector. The computing end is used to construct a multi-dimensional process state space based on the historical process feature vector, and to define a defect-free operation region and a known defect region in the multi-dimensional process state space; The current process feature vector is acquired in real time as the current process state point, and the distance from the current process state point to the boundary of the known defect area is calculated to obtain the health margin. The adjustment terminal is used to set warning boundaries. When the health margin continuously touches the warning boundaries, a risk warning is triggered. Based on the risk direction indicated by the risk warning, parameter adjustment suggestions are generated and issued.

[0017] This application discloses a method and system for processing laser equipment operation data in the Industrial Internet. By collecting and processing the laser equipment's operation information, processed material properties, optical characteristics of the cutting area, and acoustic characteristics of the cutting area, a historical process feature vector is obtained. This integrates multi-source heterogeneous data to comprehensively reflect the complex state of the laser cutting process. Based on the historical process feature vector, a multi-dimensional process state space is constructed, defining a defect-free operation area and a known defect area within this space. This allows the system to intuitively understand and quantify the risks under different operation states. The current process feature vector is acquired in real time as the current process state point, and the distance from the current process state point to the boundary of the known defect area is calculated to obtain a health margin. This health margin accurately quantifies the "safe distance" between the current operation state and potential defects, effectively overcoming the limitation of traditional threshold methods in identifying complex nonlinear defect patterns. When the health margin continuously reaches the warning boundary, a risk warning is triggered, achieving early and sensitive identification of potential risks and preventing further development of defects.

[0018] Finally, based on the risk direction indicated by the risk warning, parameter adjustment suggestions are generated and issued, realizing closed-loop control from risk identification to proactive intervention. This enables targeted adjustment of equipment parameters, effectively avoiding defects and significantly reducing high rework or scrap costs. In summary, the method of this application effectively solves the problem of difficulty in identifying and predicting early signs of hidden defects in existing technologies, improving the intelligence and reliability of the laser cutting process. Attached Figure Description

[0019] Figure 1 This is a flowchart of a laser equipment operation data processing method in the Industrial Internet provided by an embodiment of the present invention; Figure 2 This is a flowchart of a method for obtaining the current process feature vector and health margin in real time, as proposed in an embodiment of this application. Figure 3 This is a structural diagram of a laser equipment operation data processing system in the Industrial Internet provided in an embodiment of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Reference Figure 1 , Figure 1This is a flowchart of a laser equipment operation data processing method in the Industrial Internet provided by an embodiment of the present invention, including the following steps: S1 collects and processes the laser equipment's operating information, the properties of the processed material, the optical characteristics of the cutting area, and the acoustic characteristics of the cutting area to obtain a historical process feature vector. S2, Based on the historical process feature vector, construct a multi-dimensional process state space, and define a defect-free operation region and a known defect region in the multi-dimensional process state space; S3, real-time acquisition of the current process feature vector as the current process state point, and calculation of the distance from the current process state point to the boundary of the known defect area to obtain the health margin; S4, set a warning boundary; when the health margin continuously reaches the warning boundary, a risk warning is triggered. S5. Based on the risk direction indicated by the risk warning, generate parameter adjustment suggestions and issue the parameter adjustment suggestions.

[0022] The "historical process feature vector" refers to a set of data that characterizes the historical cutting process state after collecting and processing information on the laser equipment's operation, the properties of the processed materials, the optical characteristics of the cutting area, and the acoustic characteristics of the cutting area. These feature vectors are multidimensional and can comprehensively reflect all aspects of the cutting process. The "multidimensional process state space" is an abstract mathematical space whose dimensions are determined by the various features in the historical process feature vector. In this space, the "defect-free operation area" represents the set of process states that, after historical data analysis, have been determined to be able to stably produce qualified products; while the "known defect area" represents the set of process states related to specific defect patterns.

[0023] The "Current Process Feature Vector" is a data set acquired and processed in real time to characterize the current state of the laser cutting process; it is also known as the "Current Process State Point." "Health Margin" is an indicator measuring the distance between the current process state point and the boundary of a known defect area; the greater the distance, the healthier the process and the lower the risk. "Warning Boundary" is a preset health margin threshold; when the health margin reaches this boundary, it indicates that potential risks are increasing. "Risk Warning" is a signal issued by the system when it detects that the health margin is continuously approaching the warning boundary, designed to alert operators or automated systems to potential quality problems. "Risk Direction" refers to the main factors or trends indicated by the risk warning that are causing the health margin to decline. "Parameter Adjustment Suggestions" are specific equipment parameter adjustment schemes generated by the system based on the risk direction to optimize the cutting process and avoid defects.

[0024] In practical implementation, the first step is to collect and process the operating information of the laser equipment, the properties of the processed material, the optical characteristics of the cutting area, and the acoustic characteristics of the cutting area to obtain a historical process feature vector. For example, the operating information of the laser equipment can include laser power, cutting speed, focal point position, and auxiliary gas pressure and type. This data can be acquired through the equipment's built-in sensors or control system interface. The properties of the processed material can include material type, thickness, surface roughness, batch number, etc., which can be obtained from the material management system or through manual input. The optical characteristics of the cutting area can be acquired using a high-speed camera or spectrometer installed near the cutting head, such as the temperature distribution of the molten pool, the morphology of the plasma plume, and the geometry of the laser spot. The acoustic characteristics of the cutting area can be acquired using a high-sensitivity microphone, such as the frequency and amplitude of the sound waves generated during the cutting process. These raw data need to be preprocessed after acquisition, such as noise reduction, filtering, and feature extraction, and finally combined to form the historical process feature vector. For example, the average laser power, the standard deviation of the cutting speed, the texture characteristics of the molten pool image, and the specific frequency energy of the cutting sound can be used as components of the historical process feature vector.

[0025] Based on the obtained historical process feature vectors, a multidimensional process state space can be constructed, within which defect-free operation regions and known defect regions can be defined. Constructing this multidimensional process state space can be achieved through various data analysis techniques, such as principal component analysis (PCA), t-distributed random neighborhood embedding (t-SNE), or self-organizing maps (SOM), which map high-dimensional historical process feature vectors to an interpretable low-dimensional space. In the constructed state space, the definitions of defect-free operation regions and known defect regions can be trained using historical data. For example, a large amount of cutting process data for known qualified and known defective products can be collected, and their corresponding historical process feature vectors can be mapped to the state space. Through cluster analysis, classification algorithms (such as support vector machines (SVM) and neural networks), or expert experience, the boundaries representing defect-free operations and known defects can be identified and delineated. For example, the space occupied by all process state points producing qualified products can be defined as the defect-free operation region, while the space occupied by all process state points producing specific defects (such as burrs, ablation, and cracks) can be defined as the known defect region. The boundaries of these regions can be linear or non-linear, depending on the distribution characteristics of the data.

[0026] Subsequently, the current process feature vector is acquired in real time as the current process state point, and the distance from this current process state point to the boundary of the known defect region is calculated to obtain the health margin. The process of acquiring the current process feature vector in real time is similar to the acquisition of historical process feature vectors, but emphasizes real-time performance. The system continuously monitors the operating information of the laser equipment, the properties of the processed material, the optical characteristics of the cutting area, and the acoustic characteristics of the cutting area, and processes them at a preset frequency (e.g., once per second or higher) to generate a current process feature vector representing the current cutting state. Once the current process state point is acquired, its distance to the boundary of the known defect region needs to be calculated. This distance can be Euclidean distance, Mahalanobis distance, or other suitable distance metrics. For example, if the boundary of the known defect region is defined by a hyperplane, the distance can be calculated as the perpendicular distance from the point to the hyperplane. If the known defect region is a complex nonlinear region, a kernel-based method or a nearest neighbor algorithm can be used to calculate the distance. This distance value is the health margin, which intuitively reflects the degree to which the current cutting process deviates from the known defect state. The larger the health margin, the further the current process is from the defect risk, and vice versa.

[0027] Next, a warning boundary is set. When the health margin consistently reaches this boundary, a risk warning is triggered. The warning boundary is a preset health margin threshold. This threshold can be set based on historical data analysis, expert experience, or risk tolerance. For example, a lower health margin value can be set as the warning boundary. When the real-time calculated health margin drops below this value and remains below it for a period of time, the system considers there to be a potential risk. The determination of whether the warning boundary is consistently reached can be achieved through methods such as sliding window averaging, continuous counting, or time integration to avoid false alarms caused by instantaneous fluctuations. Once the condition of consistently reaching the warning boundary is met, the system will trigger a risk warning. Risk warnings can take various forms, such as displaying warning messages on the user interface, issuing audible and visual alarms, sending SMS or email notifications to relevant personnel, or even automatically activating certain security protection mechanisms.

[0028] Finally, based on the risk direction indicated by the risk warning, parameter adjustment suggestions are generated and issued. When a risk warning is triggered, the system needs to further analyze the position of the current process state point in the multi-dimensional process state space and its relative relationship with the boundary of the known defect area, thereby identifying the main factors or trends leading to the decline in health margin, i.e., the risk direction. For example, if the decline in health margin is caused by both excessively high melt pool temperature and low auxiliary gas flow rate, then the risk direction points to these two parameters. Based on the identified risk direction, the system generates specific parameter adjustment suggestions. These suggestions can be adjustments for a single parameter (e.g., "increase auxiliary gas flow rate by 5%)" or coordinated adjustments for multiple parameters (e.g., "reduce laser power by 2% while increasing cutting speed by 3%). The generation of parameter adjustment suggestions can be based on a preset rule base, optimization algorithms (such as genetic algorithms, particle swarm optimization), or machine learning models (such as reinforcement learning). For example, the system can learn and generate new adjustment suggestions based on parameter adjustment cases that successfully avoided defects in historical data. The generated parameter adjustment suggestions will then be sent to the laser equipment's control system. This can be done automatically or after confirmation by the operator, thereby achieving real-time optimization of the cutting process and defect avoidance.

[0029] This application proposes a method for processing laser equipment operation data in the Industrial Internet. By integrating multi-source heterogeneous data to construct a multi-dimensional process state space and introducing the concept of health margin, it achieves early identification and warning of potential defects in the laser cutting process. Compared with traditional methods, this application can more comprehensively capture the complex physical phenomena and material properties in the cutting process, effectively identifying hidden defect precursors caused by multi-factor nonlinear coupling. By quantifying the health margin, the system can provide an intuitive and continuous risk assessment indicator, rather than a simple binary judgment. Furthermore, this application provides parameter adjustment suggestions based on risk direction, realizing a shift from passive monitoring to proactive intervention, significantly improving the intelligence level of the production process and the stability of product quality. This method not only reduces rework and scrap costs but also improves production efficiency, providing more reliable quality assurance for the manufacturing of high-value components.

[0030] In some embodiments described above, this application proposes to acquire the current process feature vector in real time as the current process state point and calculate the distance from the current process state point to the boundary of the known defect region to obtain the health margin. However, in actual laser processing, some potential, minor process anomalies or defect precursors may be difficult to capture directly using conventional feature vectors, resulting in insufficient sensitivity of the health margin to early risks, thus affecting the timeliness and accuracy of early warnings. If the above problems are not addressed, defects may have already caused some damage before being identified, or parameters may not be effectively adjusted to avoid defects. To address this, this application further proposes a more refined and sensitive health margin acquisition method, which uses active modulation of laser parameters combined with advanced analysis of multi-source data to more accurately assess the operational health status of the laser equipment.

[0031] In this regard, refer to Figure 2 , Figure 2 This is a flowchart of a method for real-time acquisition of current process feature vectors and obtaining health margins, as proposed in an embodiment of this application, including: Step S31: Periodically modulate the laser pulse frequency or the auxiliary gas flow rate; Step S32: Collect cutting head vibration data, microphone audio data, and molten pool image sequence; Step S33: Perform spectrum analysis on the vibration data of the cutting head to obtain the energy peak value near the excitation frequency or its harmonics or sub-harmonics. Step S34: Perform spectrum analysis on the microphone audio data to obtain the energy at the perturbation frequency and its subharmonics or intermodulation frequencies. Step S35: Perform inter-frame difference and local pixel brightness fluctuation spectrum analysis on the molten pool image sequence to obtain the frequency and intensity of the periodic blur pattern and the energy of a specific low-frequency band. Step S36: The obtained energy peak value, energy, frequency and amplitude are used as new feature values ​​and added to the current process feature vector to form an extended process feature vector; Step S37: Based on the extended process feature vector, recalculate the distance from the current process state point to the boundary of the known defect region to obtain the updated health margin.

[0032] Specifically, periodic modulation of the laser pulse frequency or the flow rate of the auxiliary gas refers to introducing controllable perturbation signals by actively changing the pulse frequency of the laser or the flow rate of the auxiliary gas (such as nitrogen or oxygen). The purpose is to stimulate potential physical phenomena during the cutting process through these controlled perturbations, causing them to produce detectable responses at specific frequencies, thus making it easier to identify weak signals related to defects. For example, sine waves, square waves, or swept-frequency signals can be used to modulate the laser pulse frequency or the flow rate of the auxiliary gas; the modulation frequency and amplitude can be preset according to the specific materials and processes.

[0033] The acquisition of cutting head vibration data, microphone audio data, and molten pool image sequences refers to the simultaneous acquisition of multi-physics information during the laser cutting process using highly sensitive sensors. Cutting head vibration data, obtained through accelerometers or laser vibrometers, reflects the mechanical stability of the cutting head. Microphone audio data, acquired using industrial-grade microphones, captures acoustic characteristics generated during the cutting process, such as plasma noise and airflow disturbance sounds. Molten pool image sequences, acquired through high-speed cameras, are used to monitor the morphology, size, stability, and spatter of the molten pool in real time. This multimodal data comprehensively reflects the dynamic changes of the cutting process from different dimensions.

[0034] In practical applications, performing spectral analysis on the vibration data of the cutting head to obtain energy peaks near the excitation frequency or its harmonics and sub-harmonics refers to converting the vibration signal in the time domain into a frequency domain signal using signal processing methods such as Fourier transform, and focusing on the energy concentration at the excitation frequency and its integer multiples (harmonics) or fractions (sub-harmonics) introduced during modulation. These energy peaks are usually associated with phenomena such as resonance and vibration mode changes in the cutting head or workpiece, and are important indicators for evaluating cutting stability and identifying potential mechanical faults.

[0035] Furthermore, performing spectral analysis on the microphone audio data to obtain the energy at the perturbation frequency and its subharmonics or intermodulation frequencies refers to performing frequency domain transformation on the acquired audio signal and analyzing the energy appearing at the modulation frequency (perturbation frequency) and its nonlinear response (such as subharmonics, i.e., 1 / 2, 1 / 3, etc. of the fundamental frequency, or intermodulation frequencies, i.e., the sum and difference frequencies generated by the interaction of two or more frequency components). These energies are usually closely related to phenomena such as plasma instability, airflow turbulence, and molten pool oscillation, and are important bases for judging cutting quality and defect formation.

[0036] Furthermore, performing inter-frame differencing and local pixel brightness fluctuation spectrum analysis on the molten pool image sequence to obtain the frequency and intensity of periodic blurring patterns and the energy of specific low-frequency bands refers to image processing of molten pool images captured by a high-speed camera. Inter-frame differencing can highlight regions in the image sequence that change over time, such as fluctuations at the molten pool boundary or the movement of spatter; local pixel brightness fluctuation spectrum analysis can reveal periodic variation patterns of brightness inside or on the surface of the molten pool. The frequency and intensity of periodic blurring patterns may be related to molten pool oscillations, bubble formation, or periodic changes in plasma plumes; the energy of specific low-frequency bands may reflect the overall stability of the molten pool or the formation of macroscopic defects.

[0037] Finally, the acquired energy peak value, energy, frequency, and amplitude are used as new feature values ​​and added to the current process feature vector to form an extended process feature vector. Based on the extended process feature vector, the distance from the current process state point to the boundary of the known defect region is recalculated to obtain the updated health margin. This means that the new features obtained through the above refined analysis can more comprehensively and deeply characterize the micro-dynamics and potential anomalies of the cutting process. Combining these new features with the original current process feature vector forms a higher-dimensional and more information-rich extended process feature vector, thereby enabling the health margin calculated in the multi-dimensional process state space to more accurately reflect the distance between the current process state and the known defect region, improving the ability to identify early defects and the accuracy of health status assessment.

[0038] The proposed solution introduces controllable disturbances into the cutting process by actively and periodically modulating the laser pulse frequency or auxiliary gas flow rate, causing potential defects or anomalies to produce detectable responses at specific frequencies. This active excitation amplifies or shifts previously imperceptible weak signals to easily analyzable frequency bands. Simultaneously, by synchronously acquiring multimodal data such as cutting head vibration data, microphone audio data, and molten pool image sequences, the dynamic information of the cutting process can be comprehensively captured from multiple dimensions, including mechanical, acoustic, and optical aspects.

[0039] Subsequently, refined spectral analysis, inter-frame difference analysis, and local pixel brightness fluctuation spectral analysis of these multimodal data can identify novel features related to the modulation frequency, such as energy peaks, subharmonics, intermodulation frequencies, and periodic ambiguity patterns. These novel features can more sensitively reflect microscopic physical phenomena such as plasma instability, molten pool oscillation, and airflow disturbance during the cutting process, which are often precursors to macroscopic defects. By adding these novel features to the current process feature vector to form an extended process feature vector, the current process state point in the multidimensional process state space can more accurately represent the actual process state. This allows for earlier and more accurate detection of deviations from the normal operating region when calculating the distance to the boundary of the known defect area, thus obtaining an updated and more indicative health margin.

[0040] Through the above technical solution, this application can significantly improve the early identification capability of potential defects and anomalies in the laser cutting process. Compared with the basic solution that relies solely on conventional process feature vectors for health margin calculation, this application, through active modulation and advanced analysis of multimodal data, can capture more refined and sensitive changes in process state, thereby enabling the health margin to reflect the trend of the process deviating from the defect-free operating region earlier. This not only improves the timeliness and accuracy of risk warning and effectively avoids the expansion of defects, but also provides a more reliable basis for subsequent parameter adjustments, contributing to more refined process control and higher product quality.

[0041] In some preferred embodiments, this application is implemented as follows: In the laser cutting of stainless steel of a specific thickness, to improve the health margin's sensitivity to early molten pool instability defects, the system first periodically sinusoidally modulates the laser pulse frequency at 500 Hz with an amplitude of ±10%. Simultaneously, accelerometers mounted near the cutting head collect vibration data, high-sensitivity microphones acquire audio data from the cutting area, and a high-speed camera captures a sequence of molten pool images at 5000 frames per second. During data processing, a Fast Fourier Transform (FFT) is performed on the acquired cutting head vibration data to analyze energy peaks near 500 Hz, 1000 Hz (harmonics), and 250 Hz (divisions). Spectral analysis is performed on the microphone audio data, focusing on the energy at 500 Hz and its subharmonics (e.g., 250 Hz) or intermodulation frequencies (e.g., combinations with plasma natural frequencies).

[0042] For the molten pool image sequence, inter-frame differencing is first performed to highlight the dynamic changes at the molten pool boundary. Then, local spectral analysis is conducted on the pixel brightness in the central region of the molten pool to identify the existence and intensity of periodic blurring patterns within the 10-50Hz range. These novel features, such as energy peaks, energy, frequency, and amplitude, obtained through modulation and advanced analysis, are integrated with existing process feature vectors such as laser power, cutting speed, and auxiliary gas pressure to form an extended process feature vector containing more detailed information.

[0043] Finally, based on this extended process feature vector, the distance from the current process state point to the boundary of the known defect region is recalculated in the pre-constructed multi-dimensional process state space, thus obtaining an updated health margin that is more sensitive to early molten pool instability. When this updated health margin continuously approaches the warning boundary, the system can trigger risk warnings more promptly and generate targeted parameter adjustment suggestions based on the risk direction, such as fine-tuning the laser pulse frequency or auxiliary gas flow rate, to prevent further development of molten pool instability.

[0044] In some embodiments described above, this application proposes collecting and processing the operating information of the laser equipment, the properties of the processed material, the optical characteristics of the cutting area, and the acoustic characteristics of the cutting area to obtain a historical process feature vector. However, in a real industrial internet environment, the operating information of the laser equipment, the properties of the processed material, the optical characteristics of the cutting area, and the acoustic characteristics of the cutting area may originate from different sensors or data sources, and the time synchronization between these data sources is difficult to guarantee. Direct aggregation and processing may lead to temporal misalignment of the data, thereby affecting the accuracy of the historical process feature vector, and consequently affecting the accuracy of subsequent multi-dimensional process state space construction and health margin calculation. Therefore, this application further proposes optimizing the steps for collecting the operating information of the laser equipment, the properties of the processed material, the optical characteristics of the cutting area, and the acoustic characteristics of the cutting area to obtain the historical process feature vector, to ensure the temporal accuracy and consistency of the data.

[0045] For details, please refer to Figure 3 , Figure 3 This is a flowchart of a method for obtaining a historical process feature vector provided by an embodiment of the present invention, including: S11, Construct a high-precision time synchronization network to synchronize the clocks of data acquisition devices and computing units; S12 collects laser equipment operation information, processed material properties, optical characteristics of the cutting area, and acoustic characteristics of the cutting area, and adds a timestamp; S13, Set a microsecond-level time window and aggregate the laser equipment operation information, the properties of the processed material, the optical characteristics of the cutting area, and the acoustic characteristics of the cutting area, and combine the aggregated data into the historical process feature vector; S14, monitor the timestamp deviation of the data source and dynamically adjust the correction coefficient of the timestamp; S15, apply the correction coefficient to perform time alignment on the features in the historical process feature vector.

[0046] The purpose of building a high-precision time synchronization network is to ensure that the clocks of all devices involved in data acquisition and the computing units that perform data processing remain highly consistent. This can be achieved by using Network Time Protocol (NTP), Precision Time Protocol (PTP), or other high-precision time synchronization technologies, with the aim of laying the foundation for the accurate aggregation and analysis of subsequent multi-source data.

[0047] Furthermore, when collecting information on laser equipment operation, processed material properties, optical characteristics of the cutting area, and acoustic characteristics of the cutting area, a precise timestamp is appended to each data point. This timestamp records the exact moment the data was collected or generated, and is crucial information for achieving data time alignment.

[0048] Specifically, setting a microsecond-level time window and aggregating the laser equipment's operating information, the properties of the processed material, the optical characteristics of the cutting area, and the acoustic characteristics of the cutting area refers to integrating data from different sources and of different types, but whose timestamps fall within the same microsecond-level time window. In this way, instantaneously related multidimensional data can be aggregated into a unified dataset, and the aggregated data can be combined to form the historical process feature vector, thereby creating a comprehensive description of the state of the laser cutting process within a specific microsecond time period.

[0049] Furthermore, monitoring data source timestamp deviations and dynamically adjusting the timestamp correction factor means that the system continuously monitors the difference between the timestamps attached to each data source and the master clock provided by the high-precision time synchronization network. When a significant deviation is detected, the system dynamically calculates and adjusts a correction factor according to a preset algorithm or model to compensate for this time deviation. The purpose is to ensure that the time accuracy of the data is maintained even in the event of slight clock drift in the data source.

[0050] Finally, the correction coefficients are applied to align the features in the historical process feature vector over time. This means that before or after the formation of the historical process feature vector, the timestamps of each feature are finely adjusted using dynamically adjusted correction coefficients to ensure high consistency in the time dimension. This is crucial for accurately reflecting the instantaneous correlation between various physical quantities during laser cutting.

[0051] This application's solution constructs a high-precision time synchronization network, ensuring clock consistency between data acquisition devices and computing units from the source, thus providing a reliable reference for subsequent data timestamps. By attaching precise timestamps to various collected operational information, material properties, optical characteristics, and acoustic features, each data point carries a unique identifier of its generation time. When setting microsecond-level time windows and performing data aggregation, instantaneously related multi-source data can be accurately combined to form a historical process feature vector with high temporal resolution. Furthermore, by continuously monitoring data source timestamp deviations and dynamically adjusting correction coefficients, this application can effectively address clock drift or data transmission delays that may occur in actual industrial environments, ensuring accurate time correction of data even under dynamically changing operating conditions. Therefore, applying the correction coefficients to time-align features in the historical process feature vector allows data collected by different sensors at the same time to be accurately correlated, thereby avoiding data distortion or misjudgment caused by time misalignment.

[0052] Through the above technical solutions, this application can significantly improve the accuracy and reliability of historical process feature vectors. Precise time synchronization and data alignment ensure that the constructed multidimensional process state space can more realistically reflect the actual state of the laser cutting process, thereby making the definition of defect-free operating areas and known defect areas more accurate. This directly improves the accuracy of health margin calculation, reduces the risk of false alarms or missed alarms caused by data timing issues, provides a more solid data foundation for subsequent risk warnings and parameter adjustment suggestions, and thus improves the robustness and effectiveness of the entire laser equipment operation data processing method.

[0053] This application further proposes a more refined method for monitoring timestamp deviation and dynamically adjusting correction coefficients. By introducing a nonlinear compensation function and trend analysis, the robustness and accuracy of time synchronization are improved.

[0054] The above-mentioned monitoring of data source timestamp deviations, and the dynamic adjustment of timestamp correction coefficients, specifically include: Collect the local timestamps of each data source and the timestamp of the master clock, and calculate the instantaneous deviation; When the instantaneous deviation exceeds a preset threshold, the compensation module is activated; A nonlinear compensation function is generated based on the ambient temperature fluctuation curve, equipment operating time records, and electromagnetic interference intensity. The instantaneous deviation is corrected by applying the nonlinear compensation function to obtain the corrected deviation; Perform trend analysis on the corrected deviation; The correction coefficients are adjusted based on the results of the trend analysis.

[0055] Specifically, collecting the local timestamps of each data source and the master clock's timestamps involves recording the timestamps of data generation or acquisition using local clocks deployed on each data acquisition device (e.g., sensors, controllers), and comparing them with the unified master clock in the system. The master clock is typically a high-precision, highly stable clock source, such as a Global Positioning System (GPS) timing module or a Network Time Protocol (NTP) server. Instantaneous deviation refers to the difference between the local timestamp of each data source and the master clock's timestamp at a given moment; its purpose is to quantify the current time synchronization error.

[0056] Specifically, when the instantaneous deviation exceeds a preset threshold, the compensation module is activated. The preset threshold can be determined based on the system's time synchronization accuracy requirements, historical deviation data statistics, and the tolerance of the actual application scenario; for example, it can be set to the microsecond or nanosecond level. The compensation module is a dedicated software or hardware unit for handling time deviations. Its activation signifies that the system will switch from a simple monitoring mode to an active deviation correction mode to cope with deviations exceeding the normal range.

[0057] In practical applications, a nonlinear compensation function is generated based on the ambient temperature fluctuation curve, equipment operating time records, and electromagnetic interference intensity. The ambient temperature fluctuation curve reflects the impact of temperature on crystal oscillator frequency drift; the equipment operating time records can be used to assess the cumulative impact of equipment aging on clock accuracy; and the electromagnetic interference intensity characterizes the instantaneous disturbance of the clock signal by the external electromagnetic environment. All these factors can cause timestamp deviations to exhibit nonlinear characteristics. The nonlinear compensation function can be generated through training with historical data using machine learning algorithms (such as neural networks and support vector regression), or derived by establishing a physical model. Its purpose is to more accurately simulate and predict time deviations under complex environments.

[0058] Furthermore, the instantaneous deviation is corrected using the nonlinear compensation function to obtain the corrected deviation. This means that the instantaneous deviation is no longer adjusted simply by adding or subtracting a fixed value or a linear coefficient, but is corrected by a complex function that reflects the influence of multiple factors, thereby obtaining a deviation estimate that is closer to the true value.

[0059] Furthermore, trend analysis is performed on the corrected deviation. Trend analysis can employ statistical methods, such as moving averages, exponential smoothing, or regression analysis, to identify long-term trends, periodic fluctuations, or anomalous drift patterns in the deviation. The aim is to discover potential patterns in the deviation, providing a more forward-looking basis for subsequent adjustments to the correction coefficients.

[0060] Finally, the correction coefficients are adjusted based on the results of the trend analysis. For example, if the trend analysis shows a persistent upward or downward drift in the corrected deviation, the correction coefficients can be increased or decreased accordingly to counteract this long-term trend. This dynamic adjustment mechanism enables the system to adaptively respond to the evolution of time synchronization deviations, maintaining high-precision time alignment.

[0061] This application's solution effectively addresses the insufficient accuracy of traditional time synchronization methods in complex industrial environments by introducing multi-dimensional environmental factors and a nonlinear compensation mechanism. First, by collecting local timestamps from various data sources and comparing them with the master clock's timestamp, and calculating the instantaneous deviation, the system can monitor the real-time synchronization status. When the instantaneous deviation exceeds a preset threshold, the system can promptly activate the compensation module, shifting from passive monitoring to active intervention. Crucially, by comprehensively considering factors such as ambient temperature fluctuations, equipment operating time records, and electromagnetic interference intensity, a nonlinear compensation function is generated and applied. This allows the correction of instantaneous deviations to move beyond simple linear models, enabling more accurate fitting and compensation of nonlinear, time-varying deviations caused by these complex factors. Consequently, the corrected deviation more accurately reflects the remaining time synchronization error of the system.

[0062] Furthermore, trend analysis is performed on the corrected deviations, enabling the system to identify the long-term evolution of the deviations, rather than focusing solely on instantaneous values. Based on the results of the trend analysis, the correction coefficients are dynamically adjusted, giving the time synchronization mechanism adaptive and predictive capabilities. This allows it to anticipate and offset future deviations, ensuring that the clocks of the data acquisition equipment and computing units maintain high-precision synchronization under various complex operating conditions.

[0063] Through the above technical solution, this application can significantly improve the time synchronization accuracy and robustness of laser equipment operation data processing in the Industrial Internet. Compared with the basic solution that only performs simple monitoring and dynamically adjusts the correction coefficient, this application, by introducing a nonlinear compensation function driven by environmental factors, can more effectively eliminate nonlinear timestamp deviations caused by complex factors such as temperature, equipment aging, and electromagnetic interference, thereby obtaining more accurate corrected deviations.

[0064] Furthermore, trend analysis of the corrected deviations enables the system to identify and predict long-term variation patterns, thereby achieving adaptive adjustment of the correction coefficients and avoiding accumulated errors due to long-term drift or periodic fluctuations. This ensures that the time alignment accuracy of each feature in the historical process feature vector reaches the microsecond or even nanosecond level, providing a more reliable and accurate data foundation for subsequent construction of a multi-dimensional process state space, definition of defect-free operating areas and known defect areas, and calculation of health margins. Ultimately, this improves the accuracy of risk warnings and the effectiveness of parameter adjustment recommendations, ensuring the stable and efficient operation of laser equipment and product quality.

[0065] In some embodiments described above, this application proposes a scheme to set warning boundaries and trigger risk warnings when the health margin continuously reaches these boundaries. However, in its implementation, traditional warning mechanisms may rely solely on whether the health margin reaches a single, fixed warning boundary. This approach may fail to accurately assess the true severity and urgency of risks when faced with complex and variable operating conditions in industrial production. For example, under different processing materials, equipment loads, or environmental conditions, the fluctuation patterns of the health margin and the risk implications they indicate may differ significantly. Failure to differentiate between these differences could lead to an overly sensitive warning system with frequent false alarms, or an insufficient response when a real risk occurs, thereby affecting production efficiency and product quality. Therefore, this application further proposes a more refined and adaptive risk warning triggering mechanism that comprehensively considers multi-dimensional information to achieve tiered warnings and strategy optimization.

[0066] The above-mentioned situations, where health margins consistently approach warning thresholds, trigger risk alerts, include: Obtain current operating condition information; Acquire the frequency and duration at which health margins reach warning boundaries; Adjust the warning boundary threshold based on the current operating conditions and the frequency and duration of health margins reaching the warning boundary; Based on the adjusted warning boundary thresholds, assess the severity and urgency of the risk and trigger tiered warnings; Based on the tiered early warning system, early warning information will be pushed out. Based on the early warning information and subsequent quality inspection results, optimize the warning boundary adjustment strategy.

[0067] Specifically, acquiring current operating condition information refers to the system's real-time collection and analysis of the laser equipment's current operating status parameters, such as laser power, cutting speed, auxiliary gas type and flow rate, thickness and material of the processed material, ambient temperature, and humidity. This information is used to provide context for risk assessment. Acquiring the frequency and duration of health margins reaching warning boundaries can be understood as the system continuously monitoring the relative relationship between health margins and warning boundaries, and recording the number of times health margins fall below or approach warning boundaries and the duration of each occurrence.

[0068] For example, a time window can be set to count the number of sample points and the total duration during which the health margin falls below the warning boundary. In practical applications, adjusting the warning boundary threshold based on current operating conditions and the frequency and duration of health margin breaches means that the system dynamically adjusts the warning boundary value based on the characteristics of the current operating conditions (e.g., when processing high-reflectivity materials, the tolerance for health margin fluctuations may be low) and the frequency and duration of health margin breaches. For example, if the health margin frequently and for extended periods under specific operating conditions, it may indicate a higher potential risk, and the warning boundary threshold can be appropriately tightened to provide early warning; conversely, if the fluctuations are brief and occasional, the threshold may be appropriately relaxed to avoid false alarms.

[0069] Furthermore, assessing the severity and urgency of risks and triggering tiered warnings based on the adjusted warning boundary thresholds means that the system classifies risks into multiple levels based on the adjusted warning boundaries, combined with the degree and duration of deviation from health margins and operational information. For example, different levels such as "mild warning," "moderate warning," and "emergency shutdown warning" can be set, each corresponding to different levels of risk severity and urgency. Therefore, pushing warning information based on tiered warnings means that the system sends warning notifications to relevant personnel or systems through different channels and methods according to the assessed risk level. For example, a mild warning might be sent via email or system logs, a moderate warning might be sent via SMS or an app, and an emergency shutdown warning might trigger audible and visual alarms and automatically stop equipment operation.

[0070] Finally, based on the early warning information and subsequent quality inspection results, the warning boundary adjustment strategy is optimized. This means that the system compares the issued early warning information with the actual product quality inspection results. For example, if a minor early warning eventually leads to a product defect, the system will learn and adjust the corresponding warning boundary threshold or adjustment strategy to trigger early warnings earlier or more sensitively under similar operating conditions, thereby achieving continuous optimization and adaptive learning of the early warning mechanism.

[0071] This application's solution incorporates current operating condition information and the frequency and duration of health margin breaches at warning boundaries, transforming the risk warning mechanism from a simple binary judgment into a multi-dimensional, context-aware intelligent assessment. It is precisely this comprehensive consideration of additional information that allows the warning boundary threshold to be dynamically adjusted, thus more accurately reflecting the true risk level of the current process. Through tiered warnings, the system can take differentiated response measures based on the severity and urgency of the risk, avoiding overreaction or underreaction. Furthermore, by linking warning information with subsequent quality inspection results and using this to optimize the warning boundary adjustment strategy, a closed-loop feedback mechanism is formed, enabling the warning system to continuously learn and improve, thereby constantly enhancing its accuracy and effectiveness.

[0072] Through the above technical solutions, this application significantly enhances the intelligence and adaptability of risk warning in laser equipment operation data processing methods. Compared to traditional fixed threshold warnings, this application can dynamically adjust the warning boundaries based on actual operating conditions and the characteristics of health margin fluctuations, thereby effectively reducing false alarms and missed alarms and improving the accuracy of warnings. Furthermore, the tiered warning mechanism enables the system to provide differentiated responses based on the severity of the risk, avoiding unnecessary downtime or intervention and ensuring production continuity and efficiency. By combining warning results with actual quality inspection results for strategy optimization, the warning system of this application possesses the ability to continuously learn and self-improve, further enhancing the system's robustness and reliability, ultimately contributing to improved product quality and reduced production costs.

[0073] In some preferred embodiments, a specific example is given below. Assume a laser cutting machine is cutting a stainless steel plate. First, the system acquires real-time operating information, such as a laser power of 3kW, a cutting speed of 2m / min, nitrogen as the auxiliary gas, and a plate thickness of 5mm. Next, the system detects that the health margin has briefly touched the warning boundary three times in the past 10 minutes, each time lasting no more than 5 seconds. Based on this information, the system determines that such brief touches are likely within the normal fluctuation range under the current operating conditions. Therefore, the warning boundary threshold is dynamically adjusted, for example, slightly relaxed, to avoid triggering unnecessary warnings.

[0074] However, if the system detects that under the same operating conditions, the health margin has reached the warning boundary 10 times in the past 5 minutes, and 2 of these instances last for more than 30 seconds, the system will determine that the risk level is high based on the current operating condition information (stainless steel plate cutting) and the frequency and duration of the health margin breaches. At this time, the warning boundary threshold will be tightened, triggering a "moderate warning." The warning information will be pushed to the operator and production supervisor via SMS and the APP, prompting them to pay attention to the equipment status.

[0075] If the health margin deteriorates further and remains below the warning threshold, the system may trigger an "emergency shutdown warning," automatically stopping the laser equipment and sending an emergency notification to maintenance personnel. During subsequent product quality inspections, if a "moderate warning" ultimately results in minor burrs on the cutting edge, the system will correlate this warning information with the quality inspection results and optimize the warning threshold adjustment strategy accordingly. For example, in future situations, the system may set the trigger conditions for "moderate warnings" to be more sensitive or adjust the warning threshold to be more conservative, aiming to intervene earlier before defects occur and continuously improve the accuracy and effectiveness of warnings.

[0076] In some embodiments described above, after a risk warning is triggered, parameter adjustment suggestions are generated and issued based on the risk direction indicated by the risk warning. However, in actual industrial laser cutting processes, risk warnings may be caused by a variety of complex coupled physical phenomena, and there are complex interactions between equipment parameters. Adjusting parameters solely based on simple rules or a single objective may not effectively solve the fundamental problem and could even lead to new defects or reduced production efficiency. Therefore, this application further proposes a more refined and intelligent method for generating parameter adjustment suggestions. This method aims to generate more targeted and effective parameter adjustment suggestions by deeply analyzing the physical mechanisms behind the risks and comprehensively considering multiple optimization objectives.

[0077] The step of generating parameter adjustment suggestions based on the risk direction indicated by the risk warning and issuing the parameter adjustment suggestions includes: Based on the risk direction indicated by the risk warning, identify the dominant coupled physical phenomena in the current cutting process; Based on the dominant coupled physical phenomena, determine the set of affected equipment parameters and their interactions; set multi-parameter adjustment targets, including defect avoidance priority and production efficiency constraints; Within the range of the device parameter set, parameter adjustment candidates are generated based on the interaction relationship and the multi-parameter adjustment target; By simulating the impact of the parameter adjustment candidates on the dominant coupled physical phenomenon, the overall effect of the parameter adjustment candidates on defect avoidance and production efficiency is evaluated. Select the parameter adjustment candidate with the best overall effect as the parameter adjustment suggestion, and issue the parameter adjustment suggestion.

[0078] Specifically, identifying the dominant coupled physical phenomena in the current cutting process means that after receiving a risk warning, the system will analyze the main physical processes leading to the current risk state based on the risk type and direction indicated by the warning, combined with historical data and real-time monitoring data.

[0079] For example, if the risk warning indicates abnormal cutting edge roughness, the dominant coupled physical phenomenon might be identified as molten pool instability, excessively strong plasma plume, or turbulent auxiliary gas flow field. Determining the set of affected equipment parameters and their interactions means that once the dominant coupled physical phenomenon is identified, the system further analyzes which laser equipment parameters (e.g., laser power, pulse frequency, cutting speed, auxiliary gas pressure, nozzle height, etc.) have a significant impact on this physical phenomenon and establishes an interaction model between these parameters. This model can be built based on physical principles, empirical knowledge, or machine learning methods to quantify the impact of different parameter changes on the dominant physical phenomenon and their coupling effects.

[0080] In practical applications, setting multiple parameter adjustment goals means that, to ensure the comprehensiveness and effectiveness of parameter adjustment recommendations, several optimization objectives need to be set. These objectives specifically include defect avoidance priorities and production efficiency constraints. Defect avoidance priorities aim to ensure that the adjusted parameters can eliminate or mitigate existing defect risks to the greatest extent possible, such as avoiding burn-through, burrs, and slag buildup. Production efficiency constraints require, while avoiding defects, maintaining or increasing cutting speed and reducing energy consumption as much as possible to meet the economic requirements of industrial production.

[0081] Furthermore, within the set of device parameters, generating parameter adjustment candidates based on the interaction relationships and the multi-parameter adjustment objectives means that the system, within the determined operable range of the affected device parameters, combines the interaction relationship model between parameters and the set multi-parameter adjustment objectives to generate a series of potential parameter combinations as adjustment candidates. These candidates can be generated through optimization algorithms (such as genetic algorithms, particle swarm optimization, etc.) or rule-based expert systems.

[0082] The evaluation of the combined effect of parameter adjustment candidates on defect avoidance and production efficiency by simulating their influence on the dominant coupled physical phenomena means that for each generated parameter adjustment candidate, the system uses a pre-established physical model or simulation model to simulate the impact of that parameter combination on the dominant coupled physical phenomena during the actual cutting process. For example, it simulates the stability of the molten pool and the size of the heat-affected zone under different combinations of laser power and cutting speed. Subsequently, based on the simulation results, the combined performance of each candidate in defect avoidance (e.g., reducing defect incidence) and production efficiency (e.g., increasing cutting speed and reducing energy consumption) is evaluated.

[0083] Therefore, selecting the parameter adjustment candidate with the best overall effect as the parameter adjustment suggestion and issuing the parameter adjustment suggestion means that after evaluating all candidates, the system will select the parameter combination with the best overall effect based on preset evaluation criteria (e.g., a weighted score that comprehensively considers defect avoidance priority and production efficiency constraints). This optimal parameter combination is then determined as the parameter adjustment suggestion and issued to the laser equipment through the industrial internet platform to guide the equipment in making corresponding parameter adjustments.

[0084] This application's solution, by introducing the identification of dominant coupled physical phenomena, transforms parameter adjustment from blind trial and error or simple experience-based adjustments into a deep understanding of the problem's root cause. By identifying the set of affected equipment parameters and their interactions, the potential impact of parameter adjustments can be comprehensively grasped. Furthermore, by setting multi-parameter adjustment goals such as defect avoidance priorities and production efficiency constraints, it ensures that adjustment recommendations address defects while also considering production efficiency, avoiding situations where one aspect is prioritized at the expense of another. By generating and simulating parameter adjustment candidates, the effectiveness of different adjustment schemes can be evaluated in advance, thereby selecting the optimal solution before practical application. This significantly improves the accuracy and effectiveness of parameter adjustment, effectively overcoming the limitations of traditional methods that may result in incomplete parameter adjustments, low efficiency, or the introduction of new problems.

[0085] Through the above technical solution, this application enables more precise and intelligent intervention in the operational risks of laser equipment. Compared with methods that simply adjust parameters based on the direction of risk, this application identifies the dominant coupled physical phenomena, directly addressing the core of the problem and ensuring the targeted nature of parameter adjustment recommendations. Simultaneously, by comprehensively considering multiple parameter adjustment objectives such as defect avoidance priorities and production efficiency constraints, and conducting simulation evaluations, the generated parameter adjustment recommendations not only effectively avoid defects but also maximize or optimize production efficiency, avoiding suboptimal solutions or negative impacts caused by optimizing a single objective. Therefore, this significantly improves the intelligence level of laser equipment operation data processing and the robustness of the production process in the Industrial Internet, reducing scrap rates and production costs.

[0086] In some embodiments described above, this application proposes a scheme to generate and issue parameter adjustment suggestions based on the risk direction indicated by risk warnings. However, in the actual operation of industrial internet laser equipment, due to material diversity, equipment aging, environmental changes, and complex physical coupling effects, novel or complex coupling physical phenomena may occur that are difficult to identify using traditional methods. If these potential, nonlinear coupling phenomena cannot be accurately and promptly identified, it may lead to untimely risk warnings or inaccurate parameter adjustment suggestions, thereby affecting cutting quality and production efficiency. To address this, this application further proposes an optimized scheme for identifying the dominant coupling physical phenomena in the current cutting process. By introducing techniques such as active modulation, nonlinear analysis, and higher-order statistics, it aims to reveal potential risk sources more comprehensively and deeply.

[0087] Based on the risk direction indicated by the risk warning, the dominant coupling physical phenomena in the current cutting process are identified, including: Multi-band, multi-amplitude scanning modulation of laser pulse frequency or auxiliary gas flow rate; Collect cutting head vibration data, microphone audio data, and molten pool image sequences; Nonlinear spectral response analysis is performed on the vibration data of the cutting head, the audio data of the microphone, and the image sequence of the molten pool to identify energy peaks or mode changes that occur at the harmonics, subharmonics, or intermodulation frequencies of the modulation frequency. High-order statistical feature analysis is performed on the molten pool image sequence to identify non-periodic blur or texture anomalies; The identified energy peaks, pattern changes, blurring, or texture anomalies are used as novel features; The novel features are correlated with the current process feature vector; Identify novel coupled physical phenomena that differ significantly from the responses of known normal processes; Record the characteristic combinations of the novel coupled physical phenomena and their corresponding operating condition information; The aforementioned novel coupled physical phenomenon is incorporated into the risk early warning mechanism as a new risk model.

[0088] Specifically, "multi-band, multi-amplitude scanning modulation of the laser pulse frequency or the auxiliary gas flow rate" refers to the process of actively changing the frequency of the laser pulse or the flow rate of the auxiliary gas (e.g., oxygen, nitrogen) during laser cutting, and systematically scanning across multiple frequency ranges and different amplitudes. The aim is to introduce controlled perturbations to stimulate potential, nonlinear physical responses during the cutting process, making these responses easier to detect. For example, the auxiliary gas flow rate can be sinusoidally modulated in steps of 0.5 Hz within a range of 0.1 Hz to 10 Hz, simultaneously at different amplitudes (e.g., ±5%, ±10%).

[0089] "Acquiring cutting head vibration data, microphone audio data, and molten pool image sequences" refers to using high-precision sensors to acquire multimodal data in real time during the cutting process. Cutting head vibration data can be acquired using an accelerometer, reflecting the mechanical vibration state; microphone audio data can capture acoustic characteristics during the cutting process, such as plasma noise and airflow sounds; and molten pool image sequences are acquired using a high-speed camera to observe the dynamic behavior of the molten pool, spatter patterns, and surface morphology. This data forms the basis for subsequent analysis, used to capture subtle changes caused by modulation.

[0090] "Performing nonlinear spectral response analysis on the cutting head vibration data, the microphone audio data, and the molten pool image sequence to identify energy peaks or mode changes at the harmonics, subharmonics, or intermodulation frequencies of the modulation frequency" refers to performing spectral analysis such as Fourier transform or wavelet transform on the acquired multimodal data, with particular attention to the nonlinear response related to the modulation frequency. For example, when nonlinear coupling exists in the system, even if only one modulation frequency is applied, abnormal energy peaks or specific mode changes may be observed at the harmonics (e.g., 2f, 3f), subharmonics (e.g., f / 2, f / 3), subharmonics (e.g., f / n), or intermodulation frequencies (e.g., f1+f2, f1-f2) of that modulation frequency in the data spectrum. These nonlinear responses are key indicators for identifying novel coupled physical phenomena because they typically do not appear in normal, linear cutting processes.

[0091] "Performing high-order statistical feature analysis on the molten pool image sequence to identify aperiodic blurring or texture anomalies" refers to conducting deeper statistical analysis on the molten pool image sequence beyond traditional spectral analysis. High-order statistical features, such as skewness, kurtosis, and higher-order moments, can capture the asymmetry and kurtosis of data distribution, thereby identifying aperiodic, highly random blurred regions or anomalous textures in the images. These anomalies may indicate fluid dynamic instability within the molten pool, bubble formation, or complex changes in the solidification front, and these changes may be related to underlying defect formation mechanisms.

[0092] "Using the identified energy peaks, pattern variations, ambiguities, or texture anomalies as novel features" refers to quantifying these anomalies discovered through nonlinear spectral response analysis and higher-order statistical feature analysis into new process characteristics. These novel features can more precisely describe the micro-dynamics and potential instabilities in the cutting process.

[0093] "Correlation analysis between the novel features and the current process feature vector" refers to the comprehensive analysis of these newly discovered features with the real-time acquired current process feature vector. Machine learning algorithms (such as principal component analysis, cluster analysis, or neural networks) are used to explore the interrelationships between these novel features and known process parameters, material properties, and other conventional features in order to understand their position and significance in the entire process state space.

[0094] "Identifying novel coupled physical phenomena that differ significantly from known normal process responses" refers to using the aforementioned correlation analysis to identify feature combinations that deviate significantly from the response patterns of normal, defect-free operating regions in the feature space. These significant differences indicate the existence of unforeseen physical phenomena or interactions, which may be potential sources of future defects.

[0095] "Recording the characteristic combination of the novel coupled physical phenomenon and its corresponding operating condition information" means that once a novel coupled physical phenomenon is identified, its unique characteristic combination (e.g., a specific frequency harmonic energy peak appearing at a specific modulation frequency, accompanied by abnormal molten pool texture) and the specific operating condition information at the time of occurrence (e.g., laser power, cutting speed, material type, auxiliary gas pressure, etc.) are recorded in detail.

[0096] "Incorporating the novel coupled physical phenomena as new risk patterns into the risk early warning mechanism" means adding the recorded novel coupled physical phenomena and their characteristic combinations as new risk patterns to the existing risk early warning model. This means that the system can learn and adapt to new defect patterns, thereby improving the comprehensiveness and accuracy of early warnings and enabling early identification and prevention of unknown risks.

[0097] The aforementioned scheme aims to systematically excite potential nonlinear responses during the cutting process by actively modulating the laser pulse frequency or auxiliary gas flow rate across multiple frequency bands and amplitudes. Traditional passive monitoring may struggle to capture these hidden, nonlinear coupling phenomena. By introducing controlled perturbations, even weak, potential defect precursors can be amplified and revealed. Subsequently, the acquisition of multimodal data (cutting head vibration, microphone audio, and molten pool image sequences) provides a rich source of information for further in-depth analysis.

[0098] Furthermore, nonlinear spectral response analysis of these data can identify energy peaks or mode variations at harmonics, subharmonics, or intermodulation frequencies of the modulation frequency. These nonlinear responses directly reflect the complex coupling mechanisms within the system, which are usually not obvious or absent under normal operation. Simultaneously, high-order statistical feature analysis of the molten pool image sequence can capture aperiodic blurring or texture anomalies, which are often closely related to the hydrodynamic instability and defect formation process within the molten pool. Treating these identified energy peaks, mode variations, blurring, or texture anomalies as novel features and correlating them with the current process feature vector allows for a broader understanding of the process state.

[0099] Ultimately, by identifying novel coupled physical phenomena that differ significantly from known normal process responses and recording them as new risk patterns to incorporate into the risk warning mechanism, this approach can effectively overcome the limitations of traditional methods in identifying unknown or complex defect patterns, thereby achieving early and comprehensive perception of potential risks.

[0100] Through the aforementioned technical solution, this application significantly enhances the ability to identify the dominant coupled physical phenomena during laser cutting, especially in complex, dynamically changing industrial environments and novel defect modes. Compared to basic solutions that rely solely on known patterns for identification, this solution, through active modulation and multi-dimensional nonlinear analysis, can more sensitively capture weak signals and nonlinear responses related to potential defects that are difficult to detect using traditional methods. Therefore, the system can identify novel coupled physical phenomena that significantly differ from known normal process responses and incorporate them as new risk modes into the early warning mechanism, thereby greatly enhancing the comprehensiveness and foresight of risk warnings. This not only helps to detect potential cutting defects earlier and avoid product quality problems, but also provides more accurate risk direction indications for subsequent parameter adjustments, thereby improving production efficiency and product qualification rate, and reducing scrap rate and production costs.

[0101] In some of the embodiments described above in this application, the dominant coupled physical phenomena in the current cutting process are identified based on the risk direction indicated by the risk warning. However, in its implementation, if only a preset or known physical phenomenon model is relied upon, it may not be possible to discover and respond to new coupled physical phenomena caused by complex working conditions or material changes in a timely manner, thereby resulting in potential defects not being effectively avoided.

[0102] In response, this application further proposes a method for identifying the dominant coupled physical phenomena in the current cutting process based on the risk direction indicated by the risk warning, specifically including: Multi-band, multi-amplitude scanning modulation of laser pulse frequency or auxiliary gas flow rate; Collect cutting head vibration data, microphone audio data, and molten pool image sequences; Nonlinear spectral response analysis is performed on the vibration data of the cutting head, the audio data of the microphone, and the image sequence of the molten pool to identify energy peaks or mode changes that occur at the harmonics, subharmonics, or intermodulation frequencies of the modulation frequency. High-order statistical feature analysis is performed on the molten pool image sequence to identify non-periodic blur or texture anomalies; The identified energy peaks, pattern changes, blurring, or texture anomalies are used as novel features; The novel features are correlated with the current process feature vector; Identify novel coupled physical phenomena that differ significantly from the responses of known normal processes; Record the characteristic combinations of the novel coupled physical phenomena and their corresponding operating condition information; The aforementioned novel coupled physical phenomenon is incorporated into the risk early warning mechanism as a new risk model.

[0103] Specifically, multi-band, multi-amplitude scanning modulation of the laser pulse frequency or the auxiliary gas flow rate refers to actively and systematically changing the repetition frequency of the laser pulse or the flow rate of the auxiliary gas (such as nitrogen or oxygen) during laser cutting, and this change is performed across multiple frequency ranges and amplitude intensities. The aim is to introduce controlled perturbations to stimulate potential, imperceptible physical coupling phenomena during the cutting process, thereby enabling these phenomena to be more clearly captured and analyzed in subsequent data acquisition. For example, this can be achieved by scanning the frequency in 0.5Hz steps within the range of 0.1Hz to 10Hz, and applying amplitude modulation of ±5% to ±15% at each frequency point.

[0104] The acquisition of cutting head vibration data, microphone audio data, and molten pool image sequences refers to the real-time acquisition of multimodal data during the cutting process using high-precision sensors. Cutting head vibration data reflects the mechanical coupling state between the cutting head and the workpiece; microphone audio data captures acoustic characteristics generated during cutting, such as plasma noise and airflow noise; and molten pool image sequences directly provide visual information such as molten pool morphology, spatter, and smoke. These data form the basis for analyzing the physical phenomena of the cutting process.

[0105] In practical applications, nonlinear spectral response analysis of the aforementioned cutting head vibration data, microphone audio data, and molten pool image sequences refers to using advanced signal processing techniques, such as bispectral analysis and higher-order spectral analysis, to reveal the nonlinear interactions and coupling relationships present in these data. By identifying energy peaks or mode changes at harmonics, subharmonics, or intermodulation frequencies of the modulation frequency, physical phenomena caused by modulation that are difficult to detect in traditional linear analysis can be discovered. For example, when the laser pulse frequency is modulated, if the spectrum of the molten pool image sequence shows a significant energy peak at the second harmonic of the modulation frequency, it may indicate the existence of a certain nonlinear oscillation mode within the molten pool.

[0106] Furthermore, higher-order statistical feature analysis of molten pool image sequences refers to using statistical measures such as skewness, kurtosis, and higher-order moments to describe the non-Gaussian and non-linear characteristics of the molten pool image sequences, in addition to traditional spectral analysis. By identifying aperiodic blurring or texture anomalies, irregular changes in the surface or internal structure of the molten pool can be discovered. These changes are often closely related to the early formation of defects. For example, persistent and irregular brightness fluctuations or increased texture roughness in local areas of the molten pool image may indicate molten pool instability or oxide formation.

[0107] Using the identified energy peaks, pattern variations, ambiguities, or texture anomalies as novel features means incorporating quantitative indicators extracted through nonlinear spectral response analysis and higher-order statistical feature analysis—individual indicators reflecting potential coupled physical phenomena—as new data dimensions into the analysis system. These novel features effectively complement traditional process feature vectors.

[0108] Subsequently, correlation analysis is performed between the novel features and the current process feature vector. This involves using machine learning algorithms (such as correlation analysis, principal component analysis, and cluster analysis) to explore the intrinsic relationship between these novel features and the current cutting process state. The aim is to understand how the novel features interact with known process parameters and states, thereby providing a more comprehensive characterization of the complexity of the current process.

[0109] Therefore, identifying novel coupled physical phenomena that differ significantly from known normal process responses involves comparing the performance of novel features under current operating conditions with baseline performance under normal, defect-free operating conditions to determine whether an anomaly exists. When the combination pattern or intensity of the novel features deviates statistically significantly from normal conditions, a novel coupled physical phenomenon can be considered detected.

[0110] Furthermore, recording the characteristic combinations of the novel coupled physical phenomena and their corresponding operating condition information refers to storing a detailed description of the identified novel coupled physical phenomena (including their characteristic values, frequency of occurrence, duration, etc.) as well as information such as specific equipment parameters, material properties, and environmental conditions when the phenomenon occurs. This provides valuable data for subsequent analysis, model updates, and knowledge accumulation.

[0111] Ultimately, incorporating these novel coupled physical phenomena as new risk patterns into the risk early warning mechanism means integrating these identified and recorded phenomena, as potential precursors to defects or risk indicators, into the existing risk assessment and early warning system. This implies that the system will be able to monitor and issue early warnings for these new risk patterns in real time, thereby improving the comprehensiveness and accuracy of early warnings.

[0112] This application's solution utilizes multi-band, multi-amplitude scanning modulation of the laser pulse frequency or auxiliary gas flow rate to systematically excite potential, subtle physical coupling phenomena during the cutting process. By acquiring multimodal data (cutting head vibration data, microphone audio data, and molten pool image sequences) and performing nonlinear spectral response analysis and high-order statistical characteristic analysis, it can capture energy peaks or mode variations at harmonics, subharmonics, or intermodulation frequencies of the modulation frequency, as well as non-periodic blurring or texture anomalies, which are difficult to detect with traditional linear analysis. It is precisely this combination of active excitation, multimodal data acquisition, and advanced analysis methods that enables the system to discover and record novel risk patterns that might be overlooked under conventional monitoring, incorporating them into a risk warning mechanism, thereby effectively overcoming the limitations of relying solely on models based on known physical phenomena.

[0113] Through the aforementioned technical solution, this application can proactively detect and identify novel coupled physical phenomena that may occur during complex industrial laser cutting processes, phenomena that are difficult to detect using traditional methods. This capability significantly enhances the comprehensiveness and foresight of the risk warning mechanism, enabling the system not only to cope with known defect patterns but also to promptly detect and respond to emerging and unforeseen potential defect precursors. Therefore, it can effectively avoid production losses and product quality problems caused by the failure to identify novel defect patterns, significantly improve the stability and reliability of the laser cutting process, and provide more accurate and comprehensive risk direction indications for subsequent parameter adjustments.

[0114] In response, this application further proposes a method for generating optimized parameter adjustment suggestions, wherein generating parameter adjustment suggestions based on the risk direction indicated by the risk warning and issuing the parameter adjustment suggestions includes: Based on the risk direction indicated by the risk warning, identify the dominant coupled physical phenomena in the current cutting process; Based on the dominant coupling physical phenomena, determine the set of affected device parameters and their interaction relationships; Set multi-parameter adjustment targets, including defect avoidance priority and production efficiency constraints; Within the range of the device parameter set, parameter adjustment candidates are generated based on the interaction relationship and the multi-parameter adjustment target; By simulating the impact of the parameter adjustment candidates on the dominant coupled physical phenomenon, the overall effect of the parameter adjustment candidates on defect avoidance and production efficiency is evaluated. Select the parameter adjustment candidate with the best overall effect as the parameter adjustment suggestion, and issue the parameter adjustment suggestion.

[0115] Specifically, the phrase "obtaining the type of material currently being processed, the expected service environment of the product, and the design life requirements, and obtaining the long-term performance impact factors corresponding to the equipment parameters" means that before generating parameter adjustment suggestions, the system will obtain the specific type of material currently being processed from the production management system or material database, such as stainless steel, carbon steel, and aluminum alloy. Simultaneously, it will also obtain information on the service environment the product will face in actual applications, such as high temperature, low temperature, corrosion, and vibration, as well as the product's design life requirements. Based on this information, the system will query a preset knowledge base or model to obtain long-term performance impact factors related to the laser equipment parameters (such as laser power, cutting speed, and auxiliary gas pressure). These impact factors quantify the potential impact of different parameter settings on long-term performance indicators such as material fatigue life, corrosion resistance, and mechanical strength.

[0116] The phrase "setting multi-parameter adjustment targets, including defect avoidance priority, production efficiency constraints, and long-term performance impact weights" refers to the introduction of a "long-term performance impact weight" in addition to considering avoiding current cutting defects and maintaining production efficiency when determining parameter adjustment strategies. This weight is used to balance the relationship between immediate production goals and long-term product reliability. For example, for critical components or high-value products, the long-term performance impact weight can be set higher to ensure that the adjustment plan meets immediate needs without sacrificing the long-term quality of the product.

[0117] In practical applications, the phrase "simulating the impact of the parameter adjustment candidates on the dominant coupled physical phenomena, and evaluating the comprehensive effect of the parameter adjustment candidates on defect avoidance, production efficiency, and long-term product performance in conjunction with the long-term performance impact factors" refers to the system performing virtual simulation on each generated parameter adjustment candidate. This simulation not only predicts the improvement effect of the adjustment scheme on the current dominant coupled physical phenomena (such as molten pool instability, excessive heat-affected zone, etc.), but also, in conjunction with pre-acquired long-term performance impact factors, predicts the potential impact of the adjustment scheme on the long-term performance of the product (such as fatigue life, corrosion resistance, dimensional stability, etc.). Finally, by comprehensively considering the defect avoidance effect, changes in production efficiency, and the predicted results of the product's long-term performance, each candidate is quantitatively evaluated to obtain its comprehensive effect.

[0118] This application's solution incorporates the acquisition of information on the current material type being processed, the expected product service environment and design life requirements, as well as long-term performance impact factors corresponding to equipment parameters, when generating parameter adjustment suggestions. These factors are then included as long-term performance impact weights in the multi-parameter adjustment objectives, making the parameter adjustment decision-making process more comprehensive and intelligent. Specifically, when a risk warning indicates the need for parameter adjustment, the system not only considers how to immediately eliminate defects and maintain production efficiency but also deeply analyzes the potential impact of different parameter adjustment schemes on the long-term performance of the product. Through simulation evaluation, the system can predict whether a candidate parameter adjustment, while solving the problem in the short term, will introduce new risks or accelerate product aging in the long term.

[0119] For example, if an adjustment plan effectively avoids the current defect, but the long-term performance impact factor indicates that it may lead to increased internal stress in the material, thereby reducing fatigue life, then the overall effectiveness score of the plan will be lowered. Conversely, if a plan meets immediate requirements while having a positive or neutral impact on long-term performance, its likelihood of being selected will increase. This mechanism ensures that the generated parameter adjustment recommendations are not only short-term "stopgap" solutions, but also "fundamental" strategies that take into account the entire product lifecycle.

[0120] Through the above technical solution, this application overcomes the limitations of traditional methods that focus only on immediate defect avoidance and production efficiency while neglecting long-term product performance. Specifically, by incorporating material type, service environment, design life, and long-term performance influencing factors, and setting weights for long-term performance impact, the generated parameter adjustment recommendations can better balance short-term production goals with long-term product quality and reliability. Therefore, it can not only effectively avoid defects in the current cutting process and maintain production efficiency, but more importantly, it can significantly improve the long-term performance of the product, extend its service life, and reduce later maintenance and warranty costs, thereby enhancing the product's market competitiveness and bringing higher economic benefits and user satisfaction to the enterprise.

[0121] In some preferred embodiments, a specific example is illustrated below. Suppose that during laser cutting of stainless steel sheet, the system detects that the health margin continuously approaches the warning boundary, indicating a slight risk of molten pool instability. If relying solely on the methods described above, the system might suggest slightly increasing the laser power or decreasing the cutting speed to immediately stabilize the molten pool, avoid defects, and maintain production efficiency. However, such adjustments could result in an excessively large heat-affected zone or micro-stress concentrations within the material, thereby reducing the product's resistance to corrosive environments or its fatigue life.

[0122] According to the scheme in this application, when generating parameter adjustment suggestions, the system first obtains that the current material type being processed is "304 stainless steel," the expected product service environment is "marine environment (high salt spray, humidity)," and the design life requirement is "10 years." Simultaneously, the system queries the knowledge base to obtain long-term performance influencing factors such as "excessive laser power may increase the intergranular corrosion sensitivity of 304 stainless steel" and "excessive cutting speed may lead to an expansion of the heat-affected zone, reducing fatigue strength." When setting multi-parameter adjustment targets, in addition to defect avoidance priority and production efficiency constraints, a higher "long-term performance impact weight" is also set to emphasize the product's corrosion resistance and long-term fatigue life in a marine environment.

[0123] refer to Figure 3 , Figure 3 This application provides a structural diagram of a laser equipment operation data processing system in the Industrial Internet, comprising: The acquisition end is used to collect and process the operating information of the laser equipment, the properties of the processed material, the optical characteristics of the cutting area, and the acoustic characteristics of the cutting area to obtain the historical process feature vector. The computing end is used to construct a multi-dimensional process state space based on the historical process feature vector, and define a defect-free operation region and a known defect region in the multi-dimensional process state space; the current process feature vector is obtained in real time as the current process state point, and the distance from the current process state point to the boundary of the known defect region is calculated to obtain the health margin; The adjustment terminal is used to set warning boundaries. When the health margin continuously touches the warning boundaries, a risk warning is triggered. Based on the risk direction indicated by the risk warning, parameter adjustment suggestions are generated and issued.

[0124] This system aims to address the difficulty in identifying and predicting hidden early defects in traditional laser cutting processes. By modularizing functions such as data acquisition, state space construction, health margin calculation, risk warning, and parameter adjustment suggestion generation into independent acquisition, calculation, and adjustment ends, the system enables comprehensive, real-time monitoring and intelligent intervention of the laser equipment's operating status. The acquisition end is responsible for aggregating and preprocessing multi-source heterogeneous data, providing a foundation for subsequent analysis; the calculation end constructs an abstract process state space based on this data and quantifies the current state's health level; the adjustment end issues timely warnings and provides actionable parameter adjustment schemes based on the health assessment results. This modular system design ensures efficient collaboration among functions, thereby significantly improving the intelligence level of the laser cutting process and the stability of product quality.

[0125] The core of the laser equipment operation data processing system in the industrial internet proposed in this application lies in achieving refined management of the laser cutting process through the collaborative work of functional modules.

[0126] Specifically, this system includes a data acquisition unit. The acquisition unit is configured to acquire operating information of the laser equipment, properties of the processed material, optical characteristics of the cutting area, and acoustic characteristics of the cutting area, and processes this data to ultimately obtain a historical process feature vector.

[0127] The acquisition and processing of historical process feature vectors have been described in detail in the method embodiments. The acquisition end in this system is designed to perform the corresponding functions. It is important to emphasize that the acquisition end can be implemented in various ways at the system level. For example, the acquisition end can consist of a series of physical sensors (such as temperature sensors, pressure sensors, flow meters, high-speed cameras, spectrometers, high-sensitivity microphones, etc.), which are directly deployed on the laser equipment and cutting area, and transmit the raw data to an edge computing unit via wired or wireless networks. This edge computing unit is responsible for preliminary cleaning, noise reduction, feature extraction, and format conversion of the raw data to form historical process feature vectors. As another implementation method… The data acquisition terminal can also interface with the laser equipment's PLC / DCS system, Manufacturing Execution System (MES), and Material Management System to obtain equipment operating parameters and material attribute data. Simultaneously, it acquires optical and acoustic feature data of the cutting area through independent vision and acoustic sensors. This data is then aggregated into a central data processing server for unified processing, generating historical process feature vectors. In some implementations, to ensure the real-time nature and accuracy of data acquisition, the acquisition terminal may require manual configuration of the acquisition frequency and data format for each data source, which can introduce management complexity when there are numerous and heterogeneous data sources.

[0128] Furthermore, this system also includes a computing terminal. The computing terminal is configured to construct a multi-dimensional process state space based on the historical process feature vectors, and define a defect-free operating region and a known defect region within the multi-dimensional process state space; simultaneously, it acquires the current process feature vector in real time as the current process state point, and calculates the distance from the current process state point to the boundary of the known defect region to obtain the health margin. The specific processes of constructing the multi-dimensional process state space, defining the defect-free operating region and the known defect region, and acquiring the current process feature vector in real time and calculating the health margin have been described in detail in the method embodiments. The computing terminal in this system is designed to perform the corresponding functions.

[0129] It is important to emphasize that the computing end, at the system level, can be a high-performance server cluster running data analysis and machine learning models. For example, the computing end can deploy algorithm modules for dimensionality reduction analysis (such as Principal Component Analysis (PCA), t-distributed random neighborhood embedding (t-SNE)) and clustering analysis to construct and visualize a multi-dimensional process state space. The definitions of defect-free operating regions and known defective regions can be trained and updated using supervised learning models (such as Support Vector Machines (SVM) and neural networks). The ability to acquire current process feature vectors in real time can be implemented through a stream processing engine that continuously receives data streams from the acquisition end and calculates the current process state point in real time. The health margin calculation module can employ a geometric distance algorithm or a probability density estimation-based method. Alternatively, the computing end can also be a cloud platform, leveraging its powerful computing resources and storage capabilities to perform offline training on large-scale historical data and deploy online inference services to process current data and calculate health margins in real time. However, in some industrial settings, uploading all data to the cloud for processing may face challenges due to network bandwidth or data security limitations.

[0130] In addition, this system also includes an adjustment terminal. The adjustment terminal is configured to set warning boundaries; when the health margin continuously reaches the warning boundaries, a risk warning is triggered; and based on the risk direction indicated by the risk warning, parameter adjustment suggestions are generated and finally issued. The specific processes of setting warning boundaries, triggering risk warnings, and generating and issuing parameter adjustment suggestions based on risk direction have been described in detail in the method embodiments. The adjustment terminal in this system is designed to perform the corresponding functions.

[0131] It is important to emphasize that the adjustment terminal can be a decision support system at the system level, containing a risk assessment module and a parameter optimization module. The risk assessment module monitors the relationship between health margin and warning boundaries, and triggers tiered warnings based on preset logic (such as duration and frequency). Warning boundaries can be manually entered based on expert experience or automatically optimized through historical data analysis. The parameter optimization module can generate specific parameter adjustment suggestions based on rule engines, expert systems, or reinforcement learning models, according to the risk direction indicated by the risk warning. These suggestions can be displayed to the operator in text form for manual execution; alternatively, the adjustment terminal can communicate directly with the laser equipment controller via industrial Ethernet or OPC UA protocol, automatically issuing parameter adjustment commands. As a preferred implementation, the adjustment terminal can also integrate a human-machine interface, allowing the operator to confirm warning information and review and approve parameter adjustment suggestions to ensure production safety and quality. However, when facing complex and changing operating conditions, manually adjusting warning boundaries or parameter adjustment strategies may not respond in a timely manner, affecting the system's adaptability.

[0132] The laser equipment operation data processing system proposed in this application for the Industrial Internet provides a system-level solution to the problem of effectively identifying and predicting hidden early defects in the laser cutting process, which is difficult in existing technologies. Traditional systems often rely on threshold monitoring of single equipment parameters or simple linear correlation analysis, making it difficult to capture potential quality risks caused by complex nonlinear coupling of multi-source heterogeneous data. This system achieves closed-loop management from multi-source data acquisition to intelligent decision-making by clearly defining the functions of the acquisition end, calculation end, and adjustment end. The acquisition end can gather multi-dimensional data such as laser equipment operation information, processed material properties, and optical and acoustic characteristics of the cutting area, laying the foundation for comprehensive analysis. The calculation end, based on this, constructs a multi-dimensional process state space and quantifies health margin, providing a continuous and intuitive risk assessment indicator, rather than a simple pass / fail judgment. The adjustment end further combines risk warning with parameter adjustment suggestion generation, realizing the transformation from passive alarm to active intervention. Compared with existing systems, this system can identify potential defect precursors earlier and more accurately, and provide operable optimization solutions, thereby significantly improving the intelligence level of the laser cutting process, the stability of product quality, and production efficiency.

[0133] The above description is merely an embodiment of this application and is 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 processing laser equipment operation data in an industrial internet, characterized in that, include: Collect and process the operating information of the laser equipment, the properties of the processed material, the optical characteristics of the cutting area, and the acoustic characteristics of the cutting area to obtain a historical process feature vector. Based on the historical process feature vector, a multidimensional process state space is constructed, and a defect-free operation region and a known defect region are defined in the multidimensional process state space. The current process feature vector is acquired in real time as the current process state point, and the distance from the current process state point to the boundary of the known defect area is calculated to obtain the health margin. A warning boundary is set, and a risk warning is triggered when the health margin continuously reaches the warning boundary; Based on the risk direction indicated by the risk warning, parameter adjustment suggestions are generated and issued.

2. The method for processing laser equipment operation data in an industrial internet according to claim 1, characterized in that, The process of acquiring the current process feature vector in real time as the current process state point and calculating the distance from the current process state point to the boundary of the known defect region to obtain the health margin includes: Periodically modulate the laser pulse frequency or the auxiliary gas flow rate; Collect cutting head vibration data, microphone audio data, and molten pool image sequences; Spectral analysis is performed on the vibration data of the cutting head to obtain the energy peak value near the excitation frequency or its harmonics or sub-harmonics. Spectral analysis is performed on the microphone audio data to obtain the energy at the perturbation frequency and its subharmonics or intermodulation frequencies; Inter-frame difference and local pixel brightness fluctuation spectrum analysis are performed on the molten pool image sequence to obtain the frequency and intensity of the periodic blur pattern and the energy of a specific low-frequency band; The acquired energy peak value, energy, frequency, and amplitude are used as new feature values ​​and added to the current process feature vector to form an extended process feature vector; Based on the extended process feature vector, the distance from the current process state point to the boundary of the known defect region is recalculated to obtain the updated health margin.

3. The method for processing laser equipment operation data in an industrial internet according to claim 1, characterized in that, The acquisition of operating information of the laser equipment, properties of the processed material, optical characteristics of the cutting area, and acoustic characteristics of the cutting area yields a historical process feature vector, including: Construct a high-precision time synchronization network to synchronize the clocks of data acquisition devices and computing units; Collect laser equipment operation information, processed material properties, optical characteristics of the cutting area, and acoustic characteristics of the cutting area, and add timestamps; A microsecond-level time window is set, and the laser equipment operation information, the properties of the processed material, the optical characteristics of the cutting area, and the acoustic characteristics of the cutting area are aggregated. The aggregated data is then combined to form the historical process feature vector. Monitor data source timestamp deviation and dynamically adjust timestamp correction coefficients; The correction coefficients are applied to perform time alignment on the features in the historical process feature vector.

4. The method for processing laser equipment operation data in an industrial internet according to claim 3, characterized in that, The monitoring of data source timestamp deviation and the dynamic adjustment of the timestamp correction coefficient include: Collect the local timestamps of each data source and the timestamp of the master clock, and calculate the instantaneous deviation; When the instantaneous deviation exceeds a preset threshold, the compensation module is activated; A nonlinear compensation function is generated based on the ambient temperature fluctuation curve, equipment operating time records, and electromagnetic interference intensity. The instantaneous deviation is corrected by applying the nonlinear compensation function to obtain the corrected deviation; Perform trend analysis on the corrected deviation; The correction coefficients are adjusted based on the results of the trend analysis.

5. The method for processing laser equipment operation data in an industrial internet according to claim 1, characterized in that, When the health margin continuously reaches the warning boundary, a risk warning is triggered, including: Obtain current operating condition information; Obtain the frequency and duration at which the health margin reaches the warning boundary; Based on the current operating condition information and the frequency and duration of the health margin reaching the warning boundary, adjust the threshold of the warning boundary; Based on the adjusted warning boundary thresholds, assess the severity and urgency of the risk and trigger tiered warnings; Based on the aforementioned tiered early warning system, early warning information is pushed out. Based on the warning information and subsequent quality inspection results, the warning boundary adjustment strategy is optimized.

6. The method for processing laser equipment operation data in an industrial internet according to claim 1, characterized in that, The step of generating parameter adjustment suggestions based on the risk direction indicated by the risk warning and issuing the parameter adjustment suggestions includes: Based on the risk direction indicated by the risk warning, identify the dominant coupled physical phenomena in the current cutting process; Based on the dominant coupling physical phenomena, determine the set of affected device parameters and their interaction relationships; Set multi-parameter adjustment targets, including defect avoidance priority and production efficiency constraints; Within the range of the device parameter set, parameter adjustment candidates are generated based on the interaction relationship and the multi-parameter adjustment target; By simulating the impact of the parameter adjustment candidates on the dominant coupled physical phenomenon, the overall effect of the parameter adjustment candidates on defect avoidance and production efficiency is evaluated. Select the parameter adjustment candidate with the best overall effect as the parameter adjustment suggestion, and issue the parameter adjustment suggestion.

7. The method for processing laser equipment operation data in an industrial internet according to claim 6, characterized in that, The step of identifying the dominant coupled physical phenomena in the current cutting process based on the risk direction indicated by the risk warning includes: Multi-band, multi-amplitude scanning modulation of laser pulse frequency or auxiliary gas flow rate; Collect cutting head vibration data, microphone audio data, and molten pool image sequences; Nonlinear spectral response analysis is performed on the vibration data of the cutting head, the audio data of the microphone, and the image sequence of the molten pool to identify energy peaks or mode changes that occur at the harmonics, subharmonics, or intermodulation frequencies of the modulation frequency. High-order statistical feature analysis is performed on the molten pool image sequence to identify non-periodic blur or texture anomalies; Identified energy peaks, pattern changes, blurring, or texture anomalies are used as novel features; The novel features are correlated with the current process feature vector; Identify novel coupled physical phenomena that differ significantly from the responses of known normal processes; Record the characteristic combinations of the novel coupled physical phenomena and their corresponding operating condition information; The aforementioned novel coupled physical phenomenon is incorporated into the risk early warning mechanism as a new risk model.

8. The method for processing laser equipment operation data in an industrial internet according to claim 6, characterized in that, The step of identifying the dominant coupled physical phenomena in the current cutting process based on the risk direction indicated by the risk warning includes: Multi-band, multi-amplitude scanning modulation of laser pulse frequency or auxiliary gas flow rate; Collect cutting head vibration data, microphone audio data, and molten pool image sequences; Nonlinear spectral response analysis is performed on the vibration data of the cutting head, the audio data of the microphone, and the image sequence of the molten pool to identify energy peaks or mode changes that occur at the harmonics, subharmonics, or intermodulation frequencies of the modulation frequency. High-order statistical feature analysis is performed on the molten pool image sequence to identify non-periodic blur or texture anomalies; Identified energy peaks, pattern changes, blurring, or texture anomalies are used as novel features; The novel features are correlated with the current process feature vector; Identify novel coupled physical phenomena that differ significantly from the responses of known normal processes; Record the characteristic combinations of the novel coupled physical phenomena and their corresponding operating condition information; The aforementioned novel coupled physical phenomenon is incorporated into the risk early warning mechanism as a new risk model.

9. A method for processing laser equipment operation data in an industrial internet according to claim 1, characterized in that, The step of generating parameter adjustment suggestions based on the risk direction indicated by the risk warning and issuing the parameter adjustment suggestions includes: Based on the risk direction indicated by the risk warning, identify the dominant coupled physical phenomena in the current cutting process; Based on the dominant coupling physical phenomena, determine the set of affected device parameters and their interaction relationships; Obtain the type of material being processed, the expected product service environment, and the design life requirements, and obtain the long-term performance impact factors corresponding to the equipment parameters; Set multi-parameter adjustment targets, which include defect avoidance priority, production efficiency constraints, and long-term performance impact weights; Based on the interaction relationship, the multi-parameter adjustment target, and the long-term performance impact factor, a set of parameter adjustment candidates is generated. The effects of the parameter adjustment candidates on the dominant coupled physical phenomenon are simulated, and the comprehensive effects of the parameter adjustment candidates on defect avoidance, production efficiency and long-term product performance are evaluated in combination with the long-term performance impact factor. Select the parameter adjustment candidate with the best overall effect as the parameter adjustment suggestion, and issue the parameter adjustment suggestion.

10. A laser equipment operation data processing system in the Industrial Internet, characterized in that, include: The acquisition end is used to collect and process the operating information of the laser equipment, the properties of the processed material, the optical characteristics of the cutting area, and the acoustic characteristics of the cutting area to obtain the historical process feature vector. The computing end is used to construct a multi-dimensional process state space based on the historical process feature vector, and to define a defect-free operation region and a known defect region in the multi-dimensional process state space; The current process feature vector is acquired in real time as the current process state point, and the distance from the current process state point to the boundary of the known defect area is calculated to obtain the health margin. The adjustment end is used to set warning boundaries. When the health margin continuously reaches the warning boundaries, a risk warning is triggered. Based on the risk direction indicated by the risk warning, parameter adjustment suggestions are generated and issued.

Images