Scrap quality feedback control system of aluminum profile flexible machining center

Abstract

The invention belongs to the technical field of machining process monitoring and quality, and particularly relates to a scrap quality feedback control system of an aluminum profile flexible machining center, which comprises the following steps: S1, in the machining process, synchronously triggering a multi-source sensor array deployed in a scrap planing area, and acquiring multi-modal original data of scraps by taking a configurable time window as a unit; performing preprocessing and feature fusion on the data in each time window to generate a current multi-dimensional feature vector; when a new working condition is triggered, collecting continuous shaving data to generate a dynamic reference feature vector, performing intelligent matching and weighted fusion on the dynamic reference feature vector and historical reference features with similar process conditions in the expert knowledge base, forming a standard feature vector of the working condition, and updating the standard feature vector to the knowledge base; according to the method, self-adaptive optimization, abnormal early warning and safety intervention in the machining process can be achieved, so that the machining quality is improved, the service life of a tool is prolonged, equipment safety is guaranteed, and dependence on experience of operators is reduced.

Description

Technical Field

[0001] This invention belongs to the field of machining process monitoring and quality technology, specifically relating to a chip quality feedback control system for a flexible machining center for aluminum profiles. Background Technology

[0002] Aluminum profile flexible machining centers are highly integrated CNC machine tools capable of performing various processes such as sawing, drilling, milling, and tapping. During this efficient machining process, a large amount of aluminum chips are generated. Current technologies primarily focus on post-processing stages such as collection, cooling, and chip removal for chip disposal. However, the chips themselves are a direct "byproduct" of the machining process; their morphology (e.g., chip shape, size, color) and quality directly reflect the tool condition, the rationality of cutting parameters (e.g., spindle speed, feed rate, depth of cut), and the stability of the machining process.

[0003] Problems with existing technology: The setting of machining parameters heavily relies on operator experience or pre-set, static process databases. When abnormalities occur during machining, such as gradual tool wear, batch-to-batch material hardness differences, or insufficient cooling, there is a lack of effective real-time, online sensing and closed-loop feedback mechanisms. Typically, operators or monitoring systems can only detect problems by observing dimensional deviations or surface roughness deterioration of the workpiece after machining, or by passively hearing abnormal cutting noise or sensing abnormal machine tool vibration. By this time, workpiece scrap, tool damage, or even equipment damage have often occurred, leading to increased production costs and reduced efficiency. As a signal source generated instantaneously during machining and rich in process status information, the potential value of chip debris has not yet been systematically explored and utilized to achieve proactive, preventative closed-loop control of the machining process. Summary of the Invention

[0004] The purpose of this invention is to provide a chip quality feedback control system for a flexible machining center for aluminum profiles. This system can perceive the chip status in real time and from multiple dimensions. Through intelligent diagnosis and decision-making, it can achieve adaptive optimization, anomaly warning, and safety intervention in the machining process, thereby improving machining quality, extending tool life, ensuring equipment safety, and reducing reliance on operator experience.

[0005] The specific technical solution adopted by this invention is as follows: A method for chip quality feedback control in a flexible machining center for aluminum profiles includes the following steps: S1: During normal processing, multi-source sensors deployed near the chip area are synchronously triggered to acquire multimodal raw data of the chips in a configurable time window. After performing time-series alignment, wavelet denoising, and outlier removal on the multimodal data within each time window, the data are fused to generate the current multidimensional feature vector. In addition, during the first processing under each new working condition, continuous chip data within 2 minutes is collected, processed through the same process, and a dynamic benchmark feature vector under that working condition is generated. This vector is then intelligently matched and weighted with historical benchmark features with similar process conditions pre-stored in the expert knowledge base to form a standard feature vector representing the ideal processing state under that working condition. This vector is then updated in the knowledge base as a benchmark for subsequent comparisons. S2: By using the current multidimensional feature vector and real-time processing parameters, and utilizing the constructed digital twin mapping model, the nonlinear relationship between chip features, process parameters and underlying processing status is mapped, and the output is a status diagnosis vector; S3: Based on the obtained state diagnosis vector, query the built-in multi-level expert rule base, automatically generate and execute hierarchical control instructions: The hierarchical control commands include a first-level response for adaptive fine-tuning of process parameters, a second-level response for early warning and maintenance recommendations, and a third-level response for safety intervention. S4: Record the effect data after each diagnosis, decision and adjustment, and use the case as a new sample to implement an online dual learning mechanism for the digital twin mapping model and expert rule base to achieve the self-evolution of the system; In step S1, the multi-source sensor includes a high-speed industrial vision unit for acquiring the morphology and color characteristics of debris, an infrared thermal imaging unit for measuring the temperature field of debris, and an acoustic emission sensing unit for collecting debris fracture and friction signals. The multidimensional feature vector includes morphological distribution features, color features, temperature field features, and acoustic emission features.

[0006] In step S1, the conditions for determining the new working condition include: tool replacement, material batch change, activation of a new processing technology, or significant change in environmental conditions; the intelligent matching adopts an improved cosine similarity algorithm and comprehensively considers tool geometric parameters, material mechanical properties, and processing environment factors.

[0007] In step S2, the state diagnosis vector includes: Overall Deviation: Quantitatively represents the degree of overall difference between the current debris feature vector and the corresponding standard feature vector under the working condition; Anomaly Cause Probability Distribution: Outputs confidence assessments of various potential anomaly causes in probability form. Typical assessment objects include tool flank wear, tool edge micro-chipping, aluminum chips adhering to the tool, insufficient or failed cutting fluid supply, abnormally aggravated spindle vibration, and workpiece material hardness fluctuation.

[0008] In step S3, the triggering condition for the first-level response is that the overall deviation is between 15 and 30 and the confidence level of the specific anomaly is greater than 60%. The system automatically fine-tunes the spindle speed, feed rate or depth of cut within the preset range. The triggering condition for the secondary response is that the overall deviation is between 30 and 60 or the confidence level of the moderate anomaly is greater than 70%, and the system sends an early warning and specific maintenance suggestions. The triggering condition for the Level 3 response is that the overall deviation is greater than 60 or the confidence level of the serious anomaly is greater than 85%, and the system will forcibly execute a safe shutdown or emergency tool retraction action.

[0009] In step S3, the execution of hierarchical control instructions adopts a priority queue management mechanism. High-priority emergency safety instructions can interrupt the execution of low-priority instructions, and each instruction contains operation audit trace information.

[0010] In step S4, a dual learning mechanism is employed: Online incremental learning: The digital twin model is updated online in a lightweight manner every 100 valid samples accumulated. Regular batch optimization: Retrain the model daily using the full dataset and optimize the weights and thresholds of the expert rule base; The knowledge base management system permanently stores high-confidence success cases and manually reviews conflict cases, enabling the system to continuously evolve.

[0011] The chip quality feedback control system of the aluminum profile flexible machining center includes: A multi-source sensing and data acquisition module is used to synchronously acquire multimodal raw data of debris; The data processing and feature fusion module is used to preprocess and fuse features of the raw data, generate the current multidimensional feature vector, and is responsible for the construction and updating of the standard feature vector under the new working conditions. The digital twin mapping and state diagnosis module has a built-in deep learning-based digital twin mapping model, which is used to receive multi-dimensional feature vectors and processing parameters and output state diagnosis vectors. The intelligent decision-making and hierarchical control module has a built-in multi-level expert rule base, which is used to generate hierarchical control instructions based on the state diagnosis vector and send them to the machine tool control system. The knowledge management and self-evolution module is used to store historical data, cases, and models, and to drive the optimization and updating of models and rules based on feedback data.

[0012] It also includes a human-computer interaction module, which provides a 3D visualization interface, displays a real-time comparison cloud map of debris features and standard features, supports manual confirmation by operators or overriding of automatic diagnostic results, and generates system optimization reports.

[0013] The intelligent decision-making and hierarchical control module is communicatively connected to the machine tool CNC system. The hierarchical control instructions include code instructions that can be directly executed by the CNC system, as well as warning and guidance information sent to the human-machine interface.

[0014] The technical effects achieved by this invention are as follows: This invention analyzes "chips," the most direct and abundant byproduct of the machining process, and comprehensively utilizes their morphology, color, temperature, and acoustic emission signals to construct a multi-dimensional feature vector. This enables the system to keenly detect early warning signs of abnormalities such as tool wear, cooling failure, and material fluctuations before direct consequences such as workpiece dimensional deviations or surface quality deterioration occur, significantly shifting the monitoring window forward and transforming passive remediation into proactive prevention.

[0015] This invention automatically triggers a tiered response mechanism based on the severity of the diagnostic results. For minor deviations, it adaptively fine-tunes process parameters to optimize the process and extend tool life; for moderate anomalies, it provides timely warnings and precise maintenance suggestions to guide manual intervention; for severe risks, it immediately executes emergency interventions such as safe shutdown to protect equipment and workpieces. This layered and progressive response strategy achieves the optimal balance between production efficiency, quality assurance, and operational safety.

[0016] This invention establishes a complete data loop of "perception-decision-feedback-optimization" by recording the effect data of each control case. Utilizing the dual mechanisms of online incremental learning and periodic batch optimization, the system can continuously optimize its diagnostic model and decision rules, automatically adapting to new tooling, materials, processes, and even environmental changes. This allows the system to become increasingly intelligent with use, maintaining high precision and strong adaptability over the long term, without the need for frequent manual recalibration or major algorithm modifications.

[0017] This invention employs a deep learning-based digital twin mapping model, which can effectively learn and map the complex nonlinear relationships between multimodal features of debris, process parameters, and underlying processing states. Combined with an expert rule base, the system can not only output the overall deviation of the current state from the standard but also pinpoint the causes of anomalies probabilistically. This makes the diagnostic results not only timely but also accurate and interpretable, greatly reducing reliance on the operator's personal experience. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0019] To make the objectives and advantages of this invention clearer, the invention will be specifically described below with reference to embodiments. It should be understood that the following text is merely used to describe one or more specific embodiments of the invention and does not strictly limit the scope of protection specifically claimed by the invention.

[0020] like Figure 1 As shown, the chip quality feedback control method of the aluminum profile flexible machining center includes the following steps: S1: During normal processing, a multi-source sensor array deployed near the chip area is synchronously triggered to acquire multi-modal raw data of the chips in real time within a configurable time window of 500ms-800ms. The multi-source sensor array includes spatial and temporal synchronization mechanisms to ensure the spatiotemporal consistency of data acquisition. The multi-source sensor array includes: High-speed industrial vision unit: It adopts a global shutter CMOS sensor with a resolution of ≥2048×2048 and a frame rate of ≥500fps. It is equipped with a ring LED light source and polarizing filter to acquire the morphological and color features of debris. The morphological feature extraction adopts an improved MaskR-CNN algorithm to achieve accurate segmentation of individual debris and calculation of morphological parameters. Infrared thermal imaging unit: adopts an uncooled infrared focal plane array with a resolution of ≥320×240, a temperature measurement range of -20℃ to 500℃, an accuracy of ±2℃, and a frame rate of ≥30Hz. It eliminates the influence of environmental radiation through blackbody calibration and is used to accurately measure the temperature field distribution of debris. Acoustic emission sensing unit: It adopts a wideband resonant acoustic emission sensor with a frequency response range of 20kHz-1MHz and a sensitivity of ≥65dB. It is fixed to the machine tool spindle housing by a magnetic mounting bracket. With the help of a preamplifier and an anti-aliasing filter, it is used to acquire acoustic emission signals generated by chip fracture and friction with high fidelity. The preprocessing process for the multimodal raw data within each time window is as follows: First, the time sequence of the multi-sensor data is aligned, and the dynamic time warping algorithm is used to eliminate the time offset between sensors; second, multi-scale wavelet transform is applied for noise reduction to retain effective frequency band signals; finally, outlier detection and removal are performed based on the 3σ criterion and the isolated forest algorithm; the preprocessed data is then used to generate the current multidimensional feature vector through a feature-level fusion strategy. Multidimensional feature vectors include: Morphological distribution characteristics: including histogram of debris particle size distribution, aspect ratio statistics, curling parameters and surface roughness index; Color characteristics: Parameters for evaluating the mean chromaticity, saturation distribution, and degree of oxidation based on the CIELAB color space; Temperature field characteristics: temperature gradient on debris surface, proportion of hot spot areas, and heat distribution uniformity index; Acoustic emission characteristics: spectral energy distribution, dominant frequency offset, RMS amplitude, and time-frequency domain characteristic parameters; In each new working condition, the system automatically triggers the benchmark construction mode: continuously collects continuous chip data within 2 minutes, processes it through the same preprocessing procedure, and generates the initial dynamic benchmark feature vector for that working condition. The criteria for determining the new operating condition include: Tool change: A tool change signal is detected or the cumulative processing time of the tool exceeds the preset lifespan threshold; Material batch change: The material batch number obtained through the material management system changes, or the material composition test data deviates from the standard range; New machining process implemented: Machining program version updated, or process parameter combination exceeds the historical range by ±20%; Significant changes in environmental conditions: changes in workshop ambient temperature > ±5℃, or changes in humidity > ±15%RH; The initial vector is intelligently matched with historical benchmark feature vectors with similar process conditions pre-stored in the expert knowledge base. The matching degree is calculated using an improved cosine similarity algorithm, which comprehensively considers tool geometry parameters, material mechanical properties, and processing environment factors. The historical benchmark feature vectors that are successfully matched are weighted and fused according to the matching degree weights to finally form a standard feature vector representing the ideal processing state of the working condition. This standard feature vector includes a confidence interval and a dynamic decay factor to characterize its timeliness and reliability, and is updated to the knowledge base as a dynamic benchmark for subsequent real-time comparison. S2: Construct a deep learning-based digital twin mapping model. This model adopts a dual-channel neural network architecture: the first channel processes the current multi-dimensional feature vector, and the second channel processes real-time machining parameters, including spindle speed, feed rate, depth of cut, coolant flow rate, tool geometry parameters, and ambient temperature and humidity. The features of the two channels interact and fuse in the middle layer, highlighting key features through an attention mechanism, and finally outputting a state diagnosis vector. Training methods for digital twin mapping models include: A semi-supervised learning strategy is adopted, combining labeled expert-annotated anomaly data with unlabeled normal production data; The model loss function is designed as a multi-task learning structure, including state classification loss, parametric regression loss, and anomaly detection loss; An adversarial training mechanism is introduced, which generates synthetic anomalous samples through generative adversarial networks to improve the model's ability to identify rare anomalies; The model deployment uses model compression technology to quantize the trained deep neural network into an 8-bit integer format to meet real-time requirements; The state diagnosis vector includes: Overall deviation: The Mahalanobis distance combined with KL divergence is used to calculate the overall difference between the current debris feature vector and the corresponding standard feature vector of the working condition. Short-term fluctuations are eliminated by exponential smoothing filtering, and a quantitative score of 0-100 is output, where 0 indicates a perfect match and 100 indicates a serious deviation. Anomaly Cause Probability Distribution: Outputs the confidence assessment of multiple potential anomaly causes in probabilistic form, forming a probability vector. P=[p1,p2,...,pn ], where ∑pᵢ=1; Typical evaluation targets include: tool flank wear, tool edge chipping, aluminum chips adhering to the tool, insufficient or ineffective cutting fluid supply, abnormally increased spindle vibration, and workpiece material hardness fluctuations. S3: Based on the state diagnosis vector, query the built-in multi-level expert rule base, which includes process parameter optimization rules, maintenance decision rules, and safety protection rules; The system automatically generates and executes hierarchical control instructions: Level 1 response, adaptive fine-tuning of process parameters: When the overall deviation is between 15-30 and the confidence level of tool wear or material fluctuation in the probability distribution of abnormal causes is greater than 60%, the system activates the adaptive control module. This module is based on a real-time optimization algorithm and automatically calculates the optimal combination of cutting parameters in reverse while ensuring machining efficiency: the spindle speed adjustment range is ±10%, the feed rate adjustment range is ±8%, and the depth of cut adjustment range is ±5%. The adjustment process adopts a gradual strategy, and each adjustment does not exceed the set threshold to avoid system oscillation. After adjustment, three time windows are continuously monitored, and after confirming that the chip characteristics converge to the standard characteristics, the optimized parameters are locked. Level 2 Response, Early Warning and Maintenance Recommendations: When the overall deviation is between 30-60, or when the confidence level of moderate anomalies such as tool sticking or insufficient cooling in the probability distribution of anomaly causes exceeds 70%, the system triggers a Level 2 response; it immediately sends parameter limit commands to the machine tool CNC control system to prevent the anomaly from worsening; simultaneously, it displays graded early warning information on the operator HMI interface, including the anomaly type, severity, location, and historical trend chart; based on the maintenance case library in the knowledge base, the system generates a specific maintenance recommendation sequence, such as cleaning adhering substances on the tool rake face, checking the coolant concentration and nozzle status, calibrating the spindle balance, etc., and provides a link to the standard operating procedure manual; Level 3 Response, Safety Intervention: When the overall deviation exceeds 60%, or the confidence level of tool breakage risk or severe vibration risk in the probability distribution of abnormal causes exceeds 85%, the system immediately executes a Level 3 response; it sends the highest priority emergency control command to the CNC system, forcibly executing a safety shutdown sequence, including emergency spindle braking, rapid Z-axis retraction to a safe height, and maximum flow flushing of the cooling system; simultaneously, it triggers the workshop-level audible and visual alarm device, and the HMI interface displays a red warning sign and fault code; the system automatically saves the complete data packet within 2 minutes before the fault occurred, including raw sensor data, machining parameters, and diagnostic results, for post-event analysis; The hierarchical control instruction execution priority management mechanism includes: Establish a five-level priority queue: emergency safety instructions, process parameter adjustment instructions, early warning information instructions, data recording instructions, and learning and update instructions; High-priority instructions can interrupt the execution of low-priority instructions, ensuring timely security responses; Each instruction includes a timestamp, execution status, and operator confirmation identifier, enabling complete operation audit traceability; When an instruction fails to execute, a degradation strategy is initiated, and backup control channels are tried in turn to ensure system reliability. S4: Establish a complete data closed-loop mechanism to record each diagnosis result, control decision and adjustment effect data. The effect data includes: the debris feature vector for five consecutive time windows after adjustment, the comprehensive deviation change curve, the processing quality test results and energy consumption indicators. Based on the recorded data, the system executes a dual learning mechanism: Online incremental learning: The digital twin mapping model is updated using an online gradient descent algorithm. A lightweight model update is performed every 100 valid samples, with a focus on optimizing the accuracy of anomaly diagnosis. Regular batch optimization: After the daily production is completed, the system automatically performs model retraining, using full historical data combined with transfer learning strategies to update the rule weights and threshold parameters in the expert rule base; The knowledge base management system automatically assesses sample quality, permanently stores high-confidence successful cases, and manually reviews and confirms low-confidence or conflicting cases to ensure continuous optimization and improved reliability of the system's knowledge base.

[0021] like Figure 2 As shown, a chip quality feedback control system for a flexible machining center for aluminum profiles includes: A multi-source sensing and data acquisition module is used to synchronously acquire multimodal raw data of debris; The data processing and feature fusion module is used to preprocess and fuse features of the raw data, generate the current multidimensional feature vector, and is responsible for the construction and updating of the standard feature vector under the new working conditions. The digital twin mapping and state diagnosis module has a built-in deep learning-based digital twin mapping model, which is used to receive multi-dimensional feature vectors and processing parameters and output state diagnosis vectors. The intelligent decision-making and hierarchical control module has a built-in multi-level expert rule base, which is used to generate hierarchical control instructions based on the state diagnosis vector and send them to the machine tool control system. The knowledge management and self-evolution module is used to store historical data, cases and models, and to drive the optimization and updating of models and rules based on feedback data; The human-computer interaction module provides a 3D visualization interface that displays a real-time comparison cloud map of debris features and standard features. It supports manual confirmation by operators or overriding of automatic diagnostic results and generates system optimization reports. The intelligent decision-making and hierarchical control module is connected to the machine tool CNC system. The hierarchical control instructions include G-code or M-code instructions that can be directly executed by the CNC system, as well as early warning and guidance information sent to the human-machine interface.

[0022] Example 1: Adaptive Compensation for Normal Tool Wear Scene description: In the milling process of a certain type of aluminum alloy door and window profile, a new carbide end mill is used; the system has completed the initial benchmark learning and established the standard feature vector under the tool-material-process combination, the chip shape is a regular C-shaped chip, the color is silver-white, and the temperature field is uniform, etc.

[0023] System working process: S1: During the processing, the multi-source sensor array continuously collects debris data. The data processing module finds that in the feature vector generated within the current time window, the morphological distribution characteristics of the debris, the appearance of a small number of long strips, irregular fractured debris, acoustic emission characteristics, and RMS amplitude have a slight upward trend and a slight deviation from the standard feature vector. S2: The digital twin mapping model receives current features and real-time parameters; after model analysis, it outputs a state diagnosis vector: the overall deviation is 22, the probability distribution of abnormal causes shows a tool back face wear confidence level of 68%, and a material hardness fluctuation confidence level of 30%; S3 Intelligent Decision and Control: Based on the rule base, this situation meets the first-level response trigger condition. The intelligent decision module generates control instructions: sending instructions to the machine tool CNC system to adaptively fine-tune process parameters: slightly reducing the spindle speed from 18000rpm to 17500rpm, and simultaneously fine-tuning the feed rate from 8m / min to 7.8m / min to reduce cutting force and thermal load; on the human-machine interface, the operator is prompted: "Initial tool wear signs have been detected. Parameters have been automatically fine-tuned for compensation. Please pay attention." S4 Learning and Evolution: After parameter adjustment, subsequent time window monitoring shows that the debris characteristics gradually return to the standard range, and the overall deviation drops below 10. The system stores this successful compensation case, including the initial deviation data, adjustment instructions, and the effect after adjustment, as a high-confidence successful sample in the knowledge base. During batch optimization on the same day, the digital twin model will strengthen the mapping relationship weight of such slight wear-speed reduction and feed reduction, and the confidence of relevant rules in the expert rule base will also increase accordingly.

[0024] Implementation results: Before the operator notices or relies on traditional timed tool changing strategies, the system proactively identifies the initial performance degradation of the tool and extends the effective stable machining time of the tool by about 15-20% through closed-loop fine-tuning of parameters. This avoids workpiece dimensional deviations or surface quality degradation that may be caused by accumulated wear, thus achieving condition-based preventive maintenance.

[0025] Example 2: Rapid Diagnosis and Early Warning of Coolant Failure Scene description: When machining a deep cavity structure, a branch nozzle of the coolant became partially blocked due to impurities, resulting in a decrease in the cooling effect of that area.

[0026] System working process: S1: The infrared thermal imaging unit was the first to detect the anomaly. The temperature field characteristics of the current debris have changed significantly. The temperature of the debris in the local area of ​​the processing point is more than 40°C higher than the standard reference, and the heat distribution uniformity index has deteriorated. At the same time, the high-speed vision unit captured that the color characteristics of the debris have changed to dark blue, and the shape has become more sticky. S2: The digital twin model integrates multi-dimensional abnormal features such as high temperature, color change, and morphological adhesion, and combines them with current process parameters to quickly output diagnostic results: the overall deviation rate rapidly rises to 45, the confidence level of insufficient or failed cutting fluid supply in the probability distribution of abnormal causes is as high as 82%, and the confidence level of aluminum chips adhering to the tool is 65%; S3: This situation triggers a Level 2 response. The system immediately sends a command to the CNC system to limit the feed rate increase to prevent further increase in cutting load due to poor cooling. A yellow warning pops up on the HMI interface, accompanied by an audible alert. The interface clearly displays: "Level 2 Warning: Suspected insufficient coolant supply. Abnormal area location: X-120, Y+45; Recommended operations: 1. Immediately check if the No. 3 cooling nozzle is clogged; 2. Clean the aluminum chips adhering to the tool." At the same time, the interface compares the current high-temperature cloud map with the standard temperature field. The system simultaneously pushes this warning and maintenance recommendations to the workshop mobile management terminal. S4 Learning and Evolution: After the operator checks and clears the nozzles according to the prompts, the processing returns to normal; the system records the complete data chain from the occurrence of the anomaly, accurate diagnosis, manual intervention to the restoration of normal operation; this case, as a typical cooling anomaly-feature association sample, is added to the model training set; the weight of the multimodal feature combination of coolant failure in the knowledge base is improved in subsequent optimizations.

[0027] Implementation results: By fusing information from multiple sensors, the system can indirectly and accurately diagnose process system faults through "chips" even before the coolant flow sensor has triggered an alarm or before the operator can directly observe them. This avoids accelerated tool thermal wear, workpiece thermal deformation, and potential built-up edge phenomena caused by insufficient cooling, minimizing unplanned downtime.

[0028] Example 3: Emergency Safety Protection for Hidden Damage to Cutting Tools Scene description: During the machining process, the cutting tool suffers hidden damage due to hard points inside the material or pre-existing microcracks, but it has not yet completely failed.

[0029] System working process: S1 Data Sensing and Comparison: The acoustic emission sensing unit first captures the abnormal high-frequency, high-energy burst signal, while the high-speed vision unit observes the fragmented and fragmented abnormal debris shape in a very short time. S2 Status Diagnosis: The digital twin model receives these sharp abnormal features and makes a judgment within milliseconds, outputting a status diagnosis vector: the overall deviation instantly spikes to 75, the confidence of the risk of blade micro-chipping / breakage in the probability distribution of abnormal causes rises rapidly to 90%, and the confidence of the abnormal increase in spindle vibration is 80%; S3 Intelligent Decision-Making and Control: This directly satisfies the highest triggering condition for Level 3 response; The intelligent decision-making module immediately sends the highest priority emergency stop and tool retraction commands to the machine tool's CNC system: M00 emergency stop, G53Z0 rapid return to safe height; the machine tool spindle emergency brake, Z-axis rapid lifting; the workshop's audible and visual alarms activate instantly, and the HMI interface displays a red emergency alarm: "Level 3 alarm! Serious tool abnormality detected, emergency stop executed! Suspected cause: tool breakage; Do not continue operation, immediately check tool integrity!" The system automatically saves the complete data package from the two minutes prior to the fault for subsequent in-depth analysis. S4 Learning and Evolution: Post-incident inspection confirmed the presence of minor chipping on the tool. This case, as an extremely important successful example of "major anomaly safety handling", has been permanently stored in the high-confidence case library. The time-frequency domain characteristic patterns of the acoustic emission signal in this case, as well as its strong correlation with the abnormal chip shape, will become the gold standard for future models to identify similar hidden risks, greatly enhancing the system's ability to defend against sudden and serious failures.

[0030] Implementation results: The system detects subtle signs that the tool is about to fail completely with a speed far exceeding that of human reaction, and decisively executes the highest level of safety intervention. It successfully avoids catastrophic consequences that could have been caused by the tool continuing to cut, such as workpiece scrap, spindle damage, or even safety accidents, and protects the safety of equipment and personnel to the greatest extent, demonstrating its core value as an "intelligent safety guardian".

[0031] The above description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.

Claims

1. A method of chip mass feedback control for a flexible machining center of aluminum profiles, characterized in that, The method comprises the following steps: S1: In the processing, the multi-source sensor array deployed in the chip removal area is triggered synchronously to obtain multi-modal raw data of the chips in a configurable time window unit; The data in each time window is preprocessed and feature fused to generate a current multi-dimensional feature vector; When a new working condition is triggered, dynamic reference feature vectors are generated by collecting continuous chip removal data, and are intelligently matched and weighted fused with historical reference features with similar process conditions in the expert knowledge base to form standard feature vectors of the working condition and update to the knowledge base; S2: Through the current multi-dimensional feature vector and real-time processing parameters, a digital twin mapping model is used to map the non-linear relationship between chip features, process parameters and underlying processing states, and output a state diagnosis vector; S3: According to the obtained state diagnosis vector, the built multi-level expert rule base is queried to automatically generate and execute hierarchical control instructions: The hierarchical control instructions include a first-level response for process parameter adaptive fine-tuning, a second-level response for early warning and maintenance suggestions, and a third-level response for safety intervention S4: Record the effect data after each diagnosis, decision and adjustment, and use the case as a new sample to perform online double learning mechanism on the digital twin mapping model and the expert rule base to realize the self-evolution of the system.

2. The method of claim 1, wherein: In the step S1, the multi-source sensor includes a high-speed industrial vision unit for acquiring chip shape and color features, an infrared thermal imaging unit for measuring chip temperature field, and an acoustic emission sensing unit for collecting chip fracture and friction signals; The multi-dimensional feature vector includes shape distribution features, color features, temperature field features and acoustic emission features.

3. The method of claim 1, wherein: In the step S1, the determination conditions of the new working condition include tool replacement, material batch change, new processing technology start or significant change of environmental conditions; the intelligent matching adopts an improved cosine similarity algorithm, and comprehensively considers the tool geometric parameters, material mechanical properties and processing environment factors.

4. The method of claim 1, wherein: In the step S2, the state diagnosis vector includes: Comprehensive deviation degree: quantitatively represents the overall difference between the current chip feature vector and the corresponding working condition standard feature vector; Abnormal reason probability distribution: outputs the confidence evaluation of multiple potential abnormal reasons in the form of probability, and the typical evaluation objects include tool relief wear, tool edge micro-collapse, aluminum chip adhesion to tool, insufficient or failure of cutting fluid supply, abnormal aggravation of spindle vibration and workpiece material hardness fluctuation.

5. The method of claim 1, wherein: In the step S3, the trigger condition of the first-level response is that the comprehensive deviation degree is between 15-30 and the specific abnormal confidence is greater than 60%, and the system automatically fine-tunes the spindle speed, feed rate or cutting depth within a preset range; The trigger condition of the second-level response is that the comprehensive deviation degree is between 30-60 or the moderate abnormal confidence is greater than 70%, and the system sends a warning and specific maintenance suggestions; The trigger condition of the third-level response is that the comprehensive deviation degree is greater than 60 or the serious abnormal confidence is greater than 85%, and the system forcibly executes safety shutdown or emergency tool retracting action.

6. The method of claim 1, wherein: The step S3, the execution of the hierarchical control instruction adopts a priority queue management mechanism, the high priority emergency safety instruction can interrupt the execution of the low priority instruction, and each instruction contains operation audit tracking information.

7. The method of claim 1, wherein: The step S4, the double learning mechanism: Online incremental learning: every 100 valid samples are accumulated, that is, the digital twin model is updated online; Periodic batch optimization: retrain the model with full data and optimize the weight and threshold of the expert rule base every day; The knowledge base management system permanently stores high confidence successful cases and manually audits conflict cases to realize the continuous evolution of the system.

8. Chip mass feedback control system for a flexible machining center for aluminum profiles for implementing the method according to any one of claims 1 to 7, characterized in that, Including: Multi-source sensing and data acquisition module for synchronous acquisition of multi-modal raw data of debris; Data processing and feature fusion module for preprocessing and feature fusion of raw data, generating current multi-dimensional feature vector, and responsible for the construction and update of standard feature vector under new working conditions; Digital twin mapping and state diagnosis module, which is built-in based on deep learning digital twin mapping model, is used to receive multi-dimensional feature vector and processing parameters, and output state diagnosis vector; Intelligent decision and hierarchical control module, which is built-in multi-level expert rule base, is used to generate hierarchical control instructions according to state diagnosis vector and send them to machine tool control system; Knowledge management and self-evolution module for storing historical data, cases and models, and optimizing and updating the model and rules based on feedback data.

9. The system of claim 8, wherein: Also includes a human-computer interaction module, which provides a three-dimensional visualization interface, displays a comparison cloud diagram of debris features and standard features in real time, supports manual confirmation or override of automatic diagnosis results, and generates a system optimization report.

10. The system of claim 8, wherein: The intelligent decision and hierarchical control module is in communication connection with the machine tool numerical control system, and the hierarchical control instruction includes code instructions that can be directly executed by the numerical control system and warning and guidance information sent to the human-computer interaction interface.

Images