Numerical control machining quality monitoring method and system based on statistical process control

Abstract

The invention discloses a numerical control machining quality monitoring method and system based on statistical process control, and relates to the technical field of intelligent manufacturing and quality control. Comprising the following steps: 1, generating a multi-modal data collaborative perception and dynamic control chart; 2, federal learning optimization of knife compensation and quality association; 3, mixing enhanced quality control analysis; 4, causal reasoning driven abnormity traceability; and 5, performing digital twin closed-loop verification. Through multi-mode collaborative perception, federated learning optimization and digital twin closed-loop verification, the abnormal response speed of numerical control machining quality monitoring is increased, the compensation precision is improved, the new tool adaptive sample size is reduced, finally, the product percent of pass reaches 99.5% or above, and the comprehensive quality cost is reduced by 42%.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent manufacturing and quality control, and particularly to a numerical control machining quality monitoring method and system based on statistical process control. Background Art

[0002] With the continuous improvement of the requirements for product quality in the manufacturing industry, traditional quality control methods can no longer meet the needs of modern production. There are some problems in the practical application of traditional SPC technology: poor real-time performance, inability to perform on-line detection, difficulty in diagnosing abnormal causes, etc. At the same time, the correlation between the workpiece quality and the tool compensation value in the numerical control machining process is not clear, and it is not possible to form guiding knowledge well during the machining process. The following technical bottlenecks exist in traditional numerical control machining quality control:

[0003] 1. The off-line detection mode leads to a lag in quality feedback and cannot realize real-time optimization of the machining process;

[0004] 2. Abnormal diagnosis relies on manual experience and lacks intelligent abnormal traceability ability;

[0005] 3. The correlation between process parameter adjustment and quality indicators is not clear, and it is difficult to achieve precise compensation;

[0006] 4. The recognition accuracy of the single control chart mode is low and cannot adapt to complex machining conditions.

[0007] Although the prior art, such as the Chinese patent with the publication number CN118735365B, proposes a method based on statistical process control, it still has deficiencies in terms of real-time performance, self-adaptability, and intelligence, especially the lack of multi-modal data fusion and dynamic optimization capabilities. Summary of the Invention

[0008] Aiming at the deficiencies in the prior art, the purpose of the present invention is to provide a numerical control machining quality monitoring method and system based on statistical process control. Through multi-modal collaborative perception, federated learning optimization, and digital twin closed-loop verification, the present invention realizes an improvement in the abnormal response speed of numerical control machining quality monitoring, an increase in compensation accuracy, and a reduction in the sample size for new tool adaptation. Finally, the product qualification rate reaches over 99.5%, and the comprehensive quality cost is reduced by 42%.

[0009] To achieve the above object, the present invention is realized through the following technical solutions: A numerical control machining quality monitoring method based on statistical process control, comprising the following steps:

[0010] 1. Multi-modal data collaborative perception and dynamic control chart generation;

[0011] 2. Federated learning optimization of the correlation between tool compensation and quality;

[0012] 3. Hybrid enhanced quality control analysis;

[0013] 4. Anomaly tracing driven by causal reasoning;

[0014] 5. Digital twin closed-loop verification.

[0015] The described step 1 first performs holographic data acquisition and deploys a five-dimensional sensing network: vibration sensors, acoustic emission sensors, cutting force sensors, infrared thermal imaging sensors, and industrial vision sensors; and real-time monitors material composition fluctuations through Laser Induced Breakdown Spectroscopy (LIBS).

[0016] The described step 1 also includes a dynamic control chart selection engine, a controller based on meta-learning, and a working condition adaptive weight module to automatically adjust the control limit sensitivity according to the machine tool load rate.

[0017] The described step 2 includes distributed quality feature extraction and dynamic compensation strategies;

[0018] The distributed quality feature extraction uses a federated learning framework to construct a cross-device tool compensation knowledge base: Local model: Each machine tool independently trains a Mean-Shift clustering model; Global aggregation: Updates the Apriori association rules through secure multi-party computation; Introduces blockchain for evidence storage to ensure data immutability; The described dynamic compensation strategy: Establishes a double-closed-loop compensation system: Inner loop: Real-time fine-tuning based on Q-learning; Outer loop: Macro optimization combined with digital twin; The compensation amount calculation formula is as follows:

[0019] where α is the current gradient term and β is the historical compensation memory term;

[0020] The described step 3 includes traditional CPK analysis, deep belief network prediction, and multi-modal anomaly recognition; The traditional CPK analysis and deep belief network prediction: Input: Equipment OEE data, environmental temperature and humidity, tool life; Output: The CPK trend curve for the next 2 hours; When the predicted CPK < 1.33, trigger preventive adjustment; The multi-modal anomaly recognition uses a CNN-SVM-Transformer hybrid model architecture and adds an adversarial generative network to augment rare fault samples.

[0021] The described step 4 uses a PSO-SVR model to enhance and improve the particle swarm algorithm: Inertia weight dynamic adjustment: ω = 0.9 → 0.4 linear attenuation; Introduces a quantum behavior mutation operator to avoid premature convergence; The parameter optimization speed is increased by 60%; And uses an interpretability analysis engine, based on the double verification of SHAP values and causal graphs, to output a three-dimensional visual traceability path map.

[0022] Step 5 mentioned above adopts virtual-physical synchronous calibration, constructs a five-dimensional twin model, and conducts pre-verification of the compensation scheme: Accelerate simulation in the virtual environment: If the CPK improvement in the simulation result is less than 15%, trigger manual review; The verified scheme is automatically sent to the physical machine tool.

[0023] A numerical control machining quality monitoring system based on statistical process control includes a multi-modal data collaborative perception and dynamic control chart generation module, a federated learning optimization module for tool compensation and quality correlation, a hybrid enhanced quality control analysis module, a causal inference-driven anomaly tracing module, and a digital twin closed-loop verification module; The multi-modal data collaborative perception and dynamic control chart generation module is responsible for real-time collecting multivariate sensing data such as vibration, temperature, and cutting force, monitoring material composition through laser-induced breakdown spectroscopy, and dynamically generating a control chart adapted to the machining state, providing a high-precision and holographic process monitoring basis for the system.

[0024] The federated learning optimization module for tool compensation and quality correlation adopts federated learning technology to construct a distributed tool compensation knowledge base. On the premise of ensuring data privacy, a dynamic correlation model between tool compensation values and machining quality is established through multi-machine collaborative learning to realize the intelligent optimization of the adaptive compensation strategy.

[0025] The hybrid enhanced quality control analysis module integrates traditional SPC statistical methods and deep neural networks, calculates the process capability index (CPK) in real time and predicts the quality trend, and combines the abnormal samples generated by GAN to improve the recognition sensitivity to complex quality anomalies.

[0026] The causal inference-driven anomaly tracing module is based on the improved PSO-SVR algorithm and SHAP interpretability analysis, accurately locates the root cause of the anomaly, generates a three-dimensional tracing path map containing quantified confidence, and provides a transparent basis for decision-making.

[0027] The digital twin closed-loop verification module accelerates the simulation verification of the compensation strategy through a high-fidelity five-dimensional virtual model (geometry / physical / behavior / rule / data), and after ensuring the effectiveness of the scheme, it is sent to the physical machine tool to form a safe closed-loop of "decision-verification-execution".

[0028] Advantages of the present invention: The present invention realizes holographic process monitoring through multi-modal data collaborative perception, constructs a distributed tool compensation knowledge base based on federated learning to ensure intelligent optimization under data security, and uses a hybrid enhanced analysis method to significantly improve the sensitivity of anomaly detection. Precise traceability based on causal inference and digital twin closed-loop verification form a complete quality control chain of "perception - decision - verification - execution", achieving three major breakthroughs compared with traditional methods: First, the anomaly response speed is increased by 10 times (≤100 ms); second, the compensation accuracy is improved by 40% (CPK≥1.8); third, the sample size for new tool adaptation is reduced by 90% (only 5 - 10 groups of data are required). Finally, the product qualification rate is increased to over 99.5%, and the comprehensive quality cost is reduced by 42%, providing a reliable quality assurance system for intelligent manufacturing. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The present invention will be described in detail below in conjunction with the drawings and specific embodiments;

[0030] Figure 1 It is the system architecture diagram of the present invention. SPECIFIC EMBODIMENTS

[0031] In order to make the technical means, creative features, achieved purposes and effects of the present invention easy to understand, the present invention will be further described below in conjunction with specific embodiments.

[0032] Refer to Figure 1 , the following technical solutions are adopted in this specific embodiment: A numerical control machining quality monitoring method based on statistical process control, including the following steps:

[0033] 1. Multi-modal data collaborative perception and dynamic control chart generation

[0034] 1.1 First, perform holographic data acquisition, deploy a five-dimensional sensing network: vibration (sampling at 50 kHz) + acoustic emission (200 kHz) + cutting force (1 kHz) + infrared thermal imaging (30 fps) + industrial vision (5 μm accuracy); and monitor the material composition fluctuations in real time through laser-induced breakdown spectroscopy (LIBS)

[0035] 1.2 Dynamic control chart selection engine

[0036] A controller based on meta-learning, adding a working condition adaptive weight module to automatically adjust the sensitivity of the control limit according to the machine tool load rate;

[0037] 2. Federated learning optimization of the association between tool compensation and quality

[0038] 2.1 Distributed quality feature extraction

[0039] Adopt a federated learning framework to construct a cross-device tool compensation knowledge base:

[0040] Local model: Each machine tool independently trains a Mean-Shift clustering model

[0041] Global aggregation: Update the Apriori association rules through secure multi-party computation (SMPC), introduce blockchain for evidence storage to ensure data immutability

[0042] 2.2 Dynamic compensation strategy

[0043] Establish a double-closed-loop compensation system:

[0044] Inner loop: Real-time fine-tuning based on Q-learning (response time < 500ms)

[0045] Outer loop: Macroscopic optimization combined with digital twin (global update every 24h)

[0046] Compensation amount calculation formula:

[0047] where α is the current gradient term and β is the historical compensation memory term;

[0048] 3. Hybrid enhanced quality control analysis

[0049] 3.1 Traditional CPK analysis and Deep Belief Network (DBN) prediction:

[0050] Input: Equipment OEE data, environmental temperature and humidity, tool life;

[0051] Output: CPK trend curve for the next 2h;

[0052] Trigger preventive adjustment when the predicted CPK < 1.33;

[0053] 3.2 Multi-modal anomaly recognition

[0054] CNN-SVM-Transformer hybrid model architecture, and add a Generative Adversarial Network (GAN) to augment rare fault samples.

[0055] 4. Causal inference-driven anomaly traceability

[0056] 4.1 Adopt the PSO-SVR model to enhance and improve the particle swarm algorithm:

[0057] Dynamic adjustment of inertia weight: ω = 0.9 → 0.4 linear decay;

[0058] Introduce a quantum behavior mutation operator to avoid premature convergence;

[0059] The parameter optimization speed is increased by 60%;

[0060] 4.2 Interpretability Analysis Engine, based on the double verification of SHAP values and causal diagrams, outputs a three-dimensional visual traceability path diagram.

[0061] 5. Digital Twin Closed-loop Verification

[0062] 5.1 Virtual-Physical Synchronization Calibration

[0063] Build a five-dimensional twin model

[0064]

[0065] 5.2 Pre-verification of Compensation Scheme

[0066] Accelerate simulation in the virtual environment (10× real-time speed):

[0067] If the CPK improvement of the simulation result is less than 15%, trigger manual review;

[0068] The verified scheme is automatically sent to the physical machine tool.

[0069] A numerical control machining quality monitoring system based on statistical process control, including a multi-modal data collaborative perception and dynamic control chart generation module 1, a federated learning optimization module 2 for tool compensation and quality correlation, a hybrid enhanced quality control analysis module 3, an anomaly traceability module 4 driven by causal reasoning, and a digital twin closed-loop verification module 5; the multi-modal data collaborative perception and dynamic control chart generation module 1 is responsible for real-time collecting multivariate sensing data such as vibration, temperature, cutting force, etc., monitoring the material composition through laser-induced breakdown spectroscopy, and dynamically generating a control chart adapted to the machining state, providing a high-precision and holographic process monitoring basis for the system. Deploy a five-dimensional sensing array at key parts of the numerical control machine tool:

[0070] 1. Headstock: Install a three-axis MEMS vibration sensor (range ±50g, bandwidth 0.1 - 20kHz);

[0071] 2. Turret: Integrate an intelligent tool holder (cutting force measurement accuracy ±0.5N);

[0072] 3. Workbench: Arrange an infrared thermal imager (temperature measurement range 0 - 600°C, spatial resolution 640×480);

[0073] 4. Machining area: Set up an industrial camera (5 million pixels, equipped with a 5μm-level laser scanner);

[0074] 5. Chip conveyor: Embed an acoustic emission sensor (resonant frequency 150kHz).

[0075] And set up edge computing nodes, each machine tool is equipped with:

[0076] Industrial-grade industrial computer (Intel i7-1185G7, 32GB RAM);

[0077] Real-time data acquisition card (NI PXIe-5172, sampling rate 2MS / s);

[0078] 5G industrial module (supporting TSN time synchronization).

[0079] The federated learning optimization module 2 related to tool compensation and quality constructs a distributed tool compensation knowledge base using federated learning technology. On the premise of ensuring data privacy, a dynamic association model between tool compensation values and machining quality is established through multi-machine collaborative learning to realize the intelligent optimization of the adaptive compensation strategy.

[0080] The hybrid enhanced quality control analysis module fuses traditional SPC statistical methods and deep neural networks, calculates the process capability index (CPK) in real time and predicts the quality trend, and combines the abnormal samples generated by GAN to improve the recognition sensitivity to complex quality abnormalities.

[0081] The causal inference-driven anomaly traceability module 4 is based on the improved PSO-SVR algorithm and SHAP interpretability analysis, accurately locates the root cause of the anomaly, generates a three-dimensional traceability path map containing quantitative confidence, and provides a transparent basis for decision-making.

[0082] The digital twin closed-loop verification module 5 accelerates the simulation verification of the compensation strategy through a high-fidelity five-dimensional virtual model (geometry / physics / behavior / rules / data), and after ensuring the effectiveness of the solution, it is sent to the physical machine tool to form a safe closed-loop of "decision-making-verification-execution".

[0083] This specific implementation method constructs a holographic perception layer through a five-dimensional sensing network (vibration / acoustic emission / cutting force / thermal imaging / vision) + LIBS material monitoring, uses federated learning to optimize the tool compensation association model to achieve cross-device knowledge sharing, combines the CNN-SVM-Transformer hybrid model and GAN enhancement to improve the anomaly recognition accuracy to 93.6%, and pre-verifies the compensation strategy based on five-dimensional digital twin (geometry / physics / behavior / rules / data). Finally, an intelligent control closed-loop with a 500ms-level real-time response and a compensation accuracy of ±0.003mm is formed, reducing the scrap rate by 72% and increasing the OEE by 29 percentage points compared with the traditional SPC method, providing an interpretable, evolvable, and highly reliable new paradigm for quality control in intelligent manufacturing.

[0084] Example 1: Machining of aerospace titanium alloy structural parts: The main problem in such machining at present is thin-wall deformation, and the qualified rate of traditional machining methods is only 82%. Now, the method of the present invention is used for machining.

[0085] The specific implementation process is as follows:

[0086] 1. The acoustic emission sensor captures an abnormal frequency band of 300 kHz;

[0087] 2. Federated learning recommends a tool compensation increment ΔC = +0.03 mm;

[0088] 3. Digital twin verification shows that the deformation amount is reduced by 58%;

[0089] 4. The qualified rate of actual processing is increased to 97%.

[0090] The economic benefits of this embodiment are as follows: the unit cost is reduced by 35%, and the tool life is extended by 28%.

[0091] This implementation mode constructs an intelligent quality monitoring system covering the entire chain of "data collection - analysis - decision - verification" through the deep integration of multiple technologies. Through actual verification, the comprehensive quality cost can be reduced by more than 40%, which is particularly suitable for the intelligent manufacturing scenarios of high-value-added parts such as aerospace and precision molds.

[0092] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification only illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A numerical control machining quality monitoring method based on statistical process control, characterized in that It includes the following steps: (1) Multi-modal data collaborative perception and dynamic control chart generation; (2) Federated learning optimization of tool compensation and quality association; (3) Hybrid enhanced quality control analysis; (4) Anomaly traceability driven by causal inference; (5) Digital twin closed-loop verification.

2. The numerical control machining quality monitoring method based on statistical process control according to claim 1, wherein In the step (1), holographic data acquisition is first carried out, and a five-dimensional sensing network is deployed: vibration sensor, acoustic emission sensor, cutting force sensor, infrared thermal imaging sensor and industrial vision sensor; and the material composition fluctuation is monitored in real time by laser-induced breakdown spectroscopy.

3. A numerical control machining quality monitoring method based on statistical process control according to claim 1, characterized in that, The step (1) also includes a dynamic control chart selection engine, a controller based on meta-learning, and a working condition adaptive weight module to automatically adjust the control limit sensitivity according to the machine tool load rate.

4. A numerical control machining quality monitoring method based on statistical process control according to claim 1, characterized in that, The step (2) includes distributed quality feature extraction and dynamic compensation strategy; the distributed quality feature extraction uses a federated learning framework to build a cross-device tool compensation knowledge base: local model: each machine tool independently trains a Mean-Shift clustering model; global aggregation: update the Apriori association rule through secure multi-party computation; introduce blockchain evidence storage to ensure data immutability; the dynamic compensation strategy: establish a double-closed-loop compensation system: inner loop: real-time fine-tuning based on Q-learning; outer loop: macroscopic optimization combined with digital twin; the compensation amount calculation formula is as follows: Where α is the current gradient term and β is the historical compensation memory term.

5. A numerical control machining quality monitoring method based on statistical process control according to claim 1, characterized in that The step (3) includes traditional CPK analysis, deep belief network prediction and multi-modal anomaly recognition; the traditional CPK analysis and deep belief network prediction: input: equipment OEE data, environmental temperature and humidity, tool life; output: CPK trend curve in the next 2h; when the predicted CPK < 1.33, trigger preventive adjustment; the multi-modal anomaly recognition uses a CNN-SVM-Transformer hybrid model architecture and adds a generative adversarial network to augment rare fault samples.

6. The numerical control machining quality monitoring method based on statistical process control according to claim 1, characterized in that The step (4) uses a PSO-SVR model to enhance and improve the particle swarm optimization algorithm: inertial weight dynamic adjustment: ω = 0.9 → 0.4 linear decay; introduce a quantum behavior mutation operator to avoid premature convergence; the parameter optimization speed is increased by 60%; and an interpretability analysis engine is used to output a three-dimensional visual traceability path map based on the double verification of SHAP values and causal diagrams.

7. A numerical control machining quality monitoring method based on statistical process control according to claim 1, characterized in that, The step (5) uses virtual-entity synchronous calibration, constructs a five-dimensional twin model, and conducts pre-verification of the compensation scheme: accelerate simulation in the virtual environment: if the CPK improvement in the simulation result < 15%, trigger manual review; the verified scheme is automatically sent to the physical machine tool.

8. A numerical control machining quality monitoring system based on statistical process control, characterized in that, It includes a multi-modal data collaborative perception and dynamic control chart generation module, a federated learning optimization module for tool compensation and quality association, a hybrid enhanced quality control analysis module, an anomaly traceability module driven by causal inference, and a digital twin closed-loop verification module; the multi-modal data collaborative perception and dynamic control chart generation module is responsible for real-time collecting multi-sensor data such as vibration, temperature, and cutting force, monitoring the material composition by laser-induced breakdown spectroscopy, and dynamically generating a control chart adapted to the processing state, providing a high-precision and holographic process monitoring basis for the system.

9. A numerical control machining quality monitoring method based on statistical process control according to claim 8, characterized in that, The federated learning optimization module related to tool compensation and quality constructs a distributed tool compensation knowledge base using federated learning technology. On the premise of ensuring data privacy, a dynamic association model between tool compensation values and machining quality is established through multi-machine collaborative learning to realize the intelligent optimization of adaptive compensation strategies. The hybrid enhanced quality control analysis module integrates traditional SPC statistical methods and deep neural networks, calculates the process capability index in real time and predicts the quality trend, and combines the abnormal samples generated by GAN to improve the recognition sensitivity to complex quality abnormalities. The causal inference-driven anomaly traceability module, based on the improved PSO-SVR algorithm and SHAP interpretability analysis, accurately locates the root cause of anomalies and generates a three-dimensional traceability path map containing quantified confidence levels to provide a transparent basis for decision-making. The digital twin closed-loop verification module accelerates the simulation verification of compensation strategies through a high-fidelity five-dimensional virtual model, and after ensuring the effectiveness of the scheme, it is sent to the physical machine tool to form a safe closed loop of "decision-making - verification - execution".

Images