An integrated intelligent manufacturing unit automatic production line management method and system

Abstract

The application provides an automatic production line management method and system of an integrated intelligent manufacturing unit, and relates to the technical field of production line management.The application collects multi-source sensor data reflecting the state of a numerical control machine tool, processes the multi-source sensor data to extract a feature vector, reconstructs the extracted feature vector, obtains an abnormal score based on the extracted feature vector and the reconstructed feature vector, triggers an alarm signal based on the abnormal score, and performs physical verification of tool failure, which can minimize the impact of failure on the production process, perform thermal drift compensation on the edge node to correct the tool path when there is no tool failure, eliminate production errors caused by thermal deformation in time, obtain the compensation value of the tool used for workpiece production in the numerical control machine tool, and update the tool compensation parameter, which can better adapt to the change of the tool state, ensure the consistency and stability of the machining quality of different batches of workpieces, and improve the effectiveness of the automatic production line management.

Description

Technical Field

[0001] This invention relates to the field of production line management technology, and in particular to an automated production line management method and system integrating intelligent manufacturing units. Background Technology

[0002] As the basic working unit of a digital factory, the intelligent manufacturing unit efficiently integrates production equipment, testing devices, logistics execution mechanisms, and industrial control systems through modular, integrated, and unified design. This supports flexible production of multiple varieties and small batches of products. For example, an automated production line for connecting rods typically consists of multiple processes such as forging, machining, cleaning, testing, assembly, and logistics transportation. In the machining process, CNC machining equipment is widely used to perform key production steps such as milling, drilling, boring, broaching, precision reaming, and grinding.

[0003] Within the intelligent manufacturing unit, the Industrial Internet enables data interconnection between devices. Distributed sensors collect various types of production process data, such as vibration, acoustic emission, current, temperature, and position, and these data are preprocessed and analyzed in real time through edge computing nodes. Data processing methods include signal filtering, envelope analysis, wavelet transform, spectrum analysis, synchronous averaging, and feature extraction, thereby obtaining characteristic parameters reflecting the production status, equipment status, and environmental status.

[0004] With the application of artificial intelligence technology in the industrial field, neural network-based state recognition, anomaly detection, and predictive analysis are widely used for equipment operation monitoring and production process quality control. For example, one-dimensional convolutional autoencoders, deep neural networks, or long short-term memory networks are used to model multi-source data of the production process to learn normal operation patterns and identify abnormal deviations.

[0005] Furthermore, thermal error compensation technology is a crucial component of the intelligent manufacturing unit (ICN) to ensure and improve production accuracy. During production, machine tools are affected by factors such as ambient temperature and spindle load, leading to thermal deformation of the machine tool structure and impacting production accuracy. The ICN uses data input from multiple temperature sensors, spindle load status, and environmental variables, combined with Kalman filtering and a feedforward compensation model, to predict thermal deformation in real time and output compensation amounts. This compensation is then transmitted to the CNC system via feedforward to correct the toolpath in real time, thereby maintaining production accuracy.

[0006] For example, Chinese Patent No. CN117539169B discloses a management method and system based on digital twins, which includes: importing historical data of industrial manufacturing into a digital twin processing model, analyzing and extracting a production line comparison information set, performing perception monitoring on the production line, analyzing the management optimization requirements of each production line, and screening abnormal production lines for optimization control prompts based on the management optimization requirements of each production line.

[0007] For example, Chinese Patent Publication No. CN118732629B discloses a method and system for intelligent recommendation of production processes based on industrial IoT information cloud sharing. This includes: acquiring and storing production data from the production line based on an industrial IoT management platform; determining whether the operating parameters of the production line equipment need adjustment based on the production data; responding to the need to adjust the operating parameters of the production line equipment by generating production process parameters and adjustment time based on the production data; generating a process adjustment instruction based on the production process parameters and adjustment time and sending it to the industrial IoT management platform; parsing the process adjustment instruction through the industrial IoT management platform; and adjusting the operating parameters of the production line equipment through the control system of the industrial IoT sensing and control platform based on the process adjustment instruction when the adjustment time arrives.

[0008] The above-mentioned technology has at least the following technical problems:

[0009] In existing technologies, traditional tool condition monitoring techniques typically rely on setting fixed, simple thresholds for single or a few physical signals such as vibration and current. However, the cutting process itself is a complex dynamic process, and signals are easily affected by various factors such as cutting parameters, batch differences in workpiece materials, and environmental vibration. Abnormal conditions such as tool wear and chipping directly affect the surface quality and dimensional accuracy of the product, and may even lead to workpiece scrap and machine tool damage. Increasing sensitivity to detect minute chipping easily generates a large number of false alarms, leading to frequent downtime; while reducing sensitivity to avoid false alarms cannot effectively warn of serious tool failures.

[0010] Furthermore, traditional error compensation models are mostly linear or simple polynomial models calibrated offline through specific experiments. Once established, they are fixed in the control system and cannot adapt to the dynamic changes in actual production conditions. For example, changes in coolant condition, fluctuations in ambient temperature, progressive wear of cutting tools, and the production of different batches of materials can all cause the error characteristics of machine tools to drift, causing the original static model to quickly become inaccurate. This may introduce new, more difficult-to-trace system errors, resulting in the inability to achieve adaptive compensation that evolves synchronously with the production process, leading to low effectiveness in automated production line management. Summary of the Invention

[0011] Therefore, embodiments of the present invention provide an automated production line management method and system integrating intelligent manufacturing units, which can improve the effectiveness of automated production line management.

[0012] The technical solution of this invention is implemented as follows:

[0013] This invention provides an automated production line management method integrating intelligent manufacturing units. The method includes: S1, during workpiece production, collecting multi-source sensor data reflecting the status of a CNC machine tool, and processing the collected multi-source sensor data to extract feature vectors; S2, reconstructing the extracted feature vectors, and obtaining an anomaly score based on the extracted feature vectors and the reconstructed feature vectors to quantify the degree of difference between the two; S3, triggering an alarm signal based on the anomaly score, executing safety control operations on the spindle of the CNC machine tool, and performing physical verification of tool failure. If a tool failure is detected, an emergency stop of the spindle is triggered; otherwise, thermal drift compensation is performed at edge nodes using the collected multi-source sensor data to correct the tool path; S4, after workpiece production is completed, obtaining the compensation value of the tool used in workpiece production on the CNC machine tool based on the workpiece size measurement data, and updating the tool compensation parameters of the CNC machine tool.

[0014] This application also provides an automated production line management system integrating intelligent manufacturing units. This system is applied to an automated production line management method integrating intelligent manufacturing units. The system includes: a production process data acquisition module for acquiring multi-source sensor data reflecting the status of CNC machine tools during workpiece production, and processing the acquired multi-source sensor data to extract feature vectors; a feature vector analysis module for reconstructing the extracted feature vectors, and obtaining an anomaly score based on the extracted feature vectors and the reconstructed feature vectors to quantify the degree of difference between them; a machine tool fault monitoring module for triggering alarm signals based on the anomaly score, executing safety control operations on the spindle of the CNC machine tool, and performing physical verification of tool faults. If a tool fault is found, an emergency stop of the spindle is triggered; otherwise, thermal drift compensation is performed at edge nodes using the acquired multi-source sensor data to correct the tool path; and a machine tool fault management module for obtaining the compensation value of the tool used in workpiece production on the CNC machine tool based on workpiece size measurement data after workpiece production is completed, and updating the tool compensation parameters of the CNC machine tool.

[0015] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0016] 1. By acquiring data from multiple sensors, comprehensive information on the operating status of CNC machine tools is obtained, avoiding the limitations of data from a single sensor. Extracting feature vectors transforms complex data into a concise and effective form, facilitating subsequent analysis and processing, and laying the foundation for accurate judgment of the CNC machine tool's status. Then, by reconstructing the feature vectors and comparing them with the original feature vectors, subtle changes in the machine tool's status can be detected more acutely. Compared to existing technologies that directly analyze raw data or simply compare features, this method significantly improves the ability to detect early signs of faults. Quantifying anomaly scores makes the judgment of the CNC machine tool's status more objective and accurate, facilitating the setting of reasonable thresholds to determine whether anomalies have occurred. Finally, based on the anomaly score, an alarm signal is triggered, executing safety control operations on the spindle of the CNC machine tool. Timely triggering of alarm signals and safety control operations ensures machine tool safety. To prevent accidents and physically verify tool failures, ensuring accurate judgment, this invention minimizes the impact of failures on the production process compared to existing technologies that may misjudge or mishandle them. Different measures are taken based on the actual situation. In the absence of tool failures, thermal drift compensation at edge nodes can correct the tool path in real time, ensuring production accuracy and improving production quality. Compared to existing technologies that centralize all data processing on a central server, resulting in significant delays in thermal drift compensation, this invention reduces data transmission distance and time, improving the real-time performance of compensation. Finally, after workpiece production is completed, precise measurements of the workpiece dimensions are taken. Tool compensation values ​​are obtained from the workpiece dimension measurement data, and parameters are updated, further improving workpiece production quality and compensating for errors caused by tool wear, thermal deformation, and other factors during production, ensuring the stability of workpiece quality in subsequent production.

[0017] 2. By comprehensively utilizing multi-source sensor data such as acoustic emission signals, vibration signals, and spindle current signals, more comprehensive and richer tool status information can be obtained, enabling all-round monitoring of tool status. This significantly improves the accuracy and reliability of fault detection, effectively avoiding misjudgments and missed judgments caused by the limitations of relying on a single signal source for tool status monitoring in existing technologies. Bandpass filtering is then applied to the acoustic emission and vibration signals, with bandpass ranges set according to different frequency characteristics, effectively filtering out low-frequency mechanical vibrations and high-frequency noise, significantly improving signal quality. The envelope extracted by Hilbert transform can accurately capture the periodic impact characteristics caused by tool chipping, providing crucial evidence for early detection of tool faults. Wavelet transform decomposes the signal in the time-frequency domain, and the extracted short-time energy and peak factor reflect the local characteristics of the signal. The spectrum and spectral entropy value obtained by fast Fourier transform measure the uniformity of the signal's energy distribution from a frequency domain perspective. The spindle current is then analyzed according to the spindle rotation cycle. The current signal is synchronously averaged to eliminate random noise and periodic interference, yielding the instantaneous rate of change of the spindle current signal. This accurately reflects the changes in spindle load and provides additional reference information for tool condition monitoring. Multi-feature fusion can more comprehensively reflect tool condition information, improving the accuracy and reliability of fault detection and reducing the possibility of false positives and false negatives. Finally, anomaly scores are defined and dynamic thresholds are set, improving the accuracy and adaptability of anomaly detection. Compared with the fixed thresholds commonly used in existing technologies, dynamic thresholds can automatically adjust the judgment criteria according to the distribution of actual data. When the anomaly score exceeds the dynamic threshold, an alarm signal is triggered in a timely manner and spindle safety control operations are executed, effectively preventing further deterioration of tool faults and ensuring the safety of the production process. By measuring tool dimensions and comparing changes in tool wear, the tool condition is accurately physically verified. Based on the verification results, corresponding measures are taken to ensure the smooth operation of the production process and the safe operation of the equipment.

[0018] 3. By considering the initial ambient temperature and machine tool standby state, an initial prediction benchmark is provided for subsequent thermal drift compensation, enabling the thermal drift compensation model to start working more closely to the actual situation and improving the prediction accuracy in the initial stage. Then, at the edge computing node, using a Kalman filter, the parameters of the thermal drift compensation model are updated based on real-time collected temperature data and spindle status data, allowing the thermal drift compensation model to adapt to changes during machine tool operation in a timely manner, improving the real-time performance and accuracy of thermal drift compensation, and reducing the impact of thermal deformation on production precision. Next, on the cloud server, a physical information neural network based on a long short-term memory network incorporates physical laws into the training process, making the model training results more consistent with actual physical conditions. Key parameters for optimizing the Kalman filter are extracted from the trained model, further improving the performance of the Kalman filter and thus enhancing the accuracy of thermal drift compensation compared to existing technologies. Traditional methods often rely heavily on data-driven models, with limited consideration of physical constraints during training. This invention, however, enables model training results to better reflect actual physical conditions, avoiding predictions that contradict physical laws. This improves the reliability and generalization ability of the thermal drift compensation model. Furthermore, by collaborating with edge nodes and cloud servers, it can respond promptly to changes in machine tool operation and leverage the powerful computing capabilities of the cloud for complex model training and optimization. Finally, compared to existing technologies that use fixed optimization algorithms and simple loss functions, this invention employs an adaptive optimizer to minimize the loss function of the physical information neural network, improving training efficiency and effectiveness. The data loss term ensures that the predicted values ​​of the thermal drift compensation model are as close as possible to the true values, improving prediction accuracy. The physical loss term guarantees that the training results of the thermal drift compensation model conform to physical laws, avoiding unrealistic predictions. Overall, this invention enhances the reliability and accuracy of the thermal drift compensation model. Attached Figure Description

[0019] Figure 1 This is a flowchart of an automated production line management method integrating intelligent manufacturing units provided by an embodiment of the present invention;

[0020] Figure 2 This is a flowchart of tool state determination provided in an embodiment of the present invention;

[0021] Figure 3 This is a model architecture diagram of the physical information neural network provided in the embodiments of the present invention;

[0022] Figure 4 This is a flowchart of data offset compensation provided in an embodiment of the present invention;

[0023] Figure 5 This is a schematic diagram of the structure of an automated production line management system integrating intelligent manufacturing units provided in an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on 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.

[0025] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0026] Example 1: This embodiment of the invention provides a method for managing an automated production line integrating intelligent manufacturing units. For example... Figure 1 The flowchart shown represents an automated production line management method integrating intelligent manufacturing units. The processing flow of this method may include the following steps:

[0027] S1. During the workpiece production process, multi-source sensor data reflecting the status of the CNC machine tool is collected synchronously, and the collected multi-source sensor data is processed to extract feature vectors. Multi-source sensor data refers to data from various types of sensors, including but not limited to temperature sensors for measuring the temperature of various parts of the machine tool, vibration sensors for monitoring machine tool vibration, displacement sensors for detecting changes in the position of the tool or workpiece, and pressure sensors for measuring pressure-related parameters such as cutting force. By collecting multi-source sensor data, comprehensive information on the machine tool's operating status is obtained, avoiding the limitations that may exist with single sensor data. By extracting feature vectors, subsequent analysis and processing are facilitated, laying the foundation for accurately judging the machine tool's status.

[0028] S2 reconstructs the extracted feature vectors. Based on the extracted and reconstructed feature vectors, an anomaly score is obtained to quantify the degree of difference between the two, i.e., the extracted and reconstructed feature vectors. By reconstructing the feature vectors and comparing them with the original feature vectors, the judgment of the machine tool status becomes more objective and accurate, making it easier to set reasonable thresholds to determine whether the machine tool has an anomaly.

[0029] S3, based on the abnormal score triggering alarm signal, executes the safety control operation of the spindle in the CNC machine tool and performs physical verification of tool failure. If tool failure is detected, the spindle is triggered to stop immediately to prevent the faulty tool from continuing to run and causing further damage to the workpiece and machine tool. Otherwise, thermal drift compensation is performed at the edge node using the collected multi-source sensor data to correct the tool path, ensure machining accuracy, and improve production quality. In this application, the edge node is a computing device installed near the CNC machine tool, which can quickly process locally collected sensor data. Timely triggering of alarm signals and safety control operations can ensure the safety of the CNC machine tool and prevent accidents. The physical verification of tool failure ensures the accuracy of the judgment.

[0030] S4. After the workpiece is produced, the compensation value of the tool used in the production of the workpiece is obtained based on the measurement data of the workpiece size, and the tool compensation parameters of the CNC machine tool are updated. The tool compensation value is obtained and the parameters are updated through the workpiece size measurement data to compensate for the errors caused by factors such as tool wear and thermal deformation during the production process, so as to ensure the stability of the workpiece quality in subsequent production.

[0031] Furthermore, the collected multi-source sensor data is processed to extract feature vectors. The specific process is as follows:

[0032] A first bandpass filter is applied to the acoustic emission signal in the multi-source sensor data, and a second bandpass filter is applied to the vibration signal in the multi-source sensor data to filter out low-frequency mechanical vibrations and high-frequency noise. For example, low-frequency mechanical vibrations may come from the machine tool itself, and high-frequency noise may come from interference from the surrounding environment. Bandpass filtering allows signals within a certain frequency range to pass through while suppressing signals outside that frequency range. The bandpass range of the first bandpass filter is larger than that of the second bandpass filter. The bandpass ranges of the first and second bandpass filters are set by preset personnel according to the characteristics of the corresponding signals. For example, a bandpass filter of 100kHz-400kHz is applied to the acoustic emission signal, and a bandpass filter of 1kHz-10kHz is applied to the vibration signal. Through bandpass filtering, low-frequency mechanical vibrations and high-frequency noise in the acoustic emission signal and vibration signal are effectively filtered out, improving the signal quality and signal-to-noise ratio, making subsequent analysis more accurate and reliable.

[0033] Hilbert transforms are applied to both the bandpass-filtered acoustic and vibration signals to extract their envelopes. This envelope is used to capture the periodic impact characteristics caused by tool chipping. The envelope is a smooth curve connecting a series of peak points in the bandpass-filtered acoustic or vibration signal waveform. The Hilbert transform is a linear time-invariant filter. For a real signal, the Hilbert-transformed signal combined with the real signal can form an analytic signal. The envelope can be obtained by taking the modulus of the analytic signal. When tool chipping occurs, periodic impact pulses are generated in the acoustic and vibration signals. The envelope can clearly show the amplitude changes of the impact pulses, thus determining whether the tool has malfunctioned.

[0034] Wavelet transform is used to decompose the bandpass-filtered signal and extract short-time energy and peak factor within a preset time window. Wavelet transform is a time-scale analysis method that decomposes the signal onto wavelet basis functions of different scales, enabling simultaneous analysis in the time and frequency domains. It can extract local features of the signal at different time intervals. Short-time energy reflects the energy of the signal within a short period, and peak factor is the ratio of the signal peak value to the effective value. Specifically, the signal after bandpass filtering is first divided into frames, and the squares of all sampled points in each frame are summed to obtain the short-time energy. The maximum absolute value of all sampled points in the frame is recorded as the peak value. The average of the squares of all sampled points in the frame is calculated, and then the square root of this average is taken to obtain the effective value. By calculating short-time energy and peak factor within a preset time window, the energy distribution and peak characteristics of the signal within a local time range can be analyzed, effectively detecting transient features and anomalies in the signal. For example, tool chipping may cause a brief energy concentration and peak change, which helps to detect tool failures in a timely manner.

[0035] Fast Fourier Transform (FFT) is performed on the bandpass-filtered acoustic emission and vibration signals to obtain the spectrum, and the spectral entropy value is then acquired. The spectral entropy value is a characteristic quantity calculated based on the spectrum. First, the amplitude of the complex value corresponding to each frequency point in the spectrum is recorded as the amplitude spectrum. The amplitude spectrum is then normalized by comparing the amplitude value at each frequency point with the sum of the entire amplitude spectrum. Then, the logarithm to base 2 is taken for each normalized amplitude value, and this normalized amplitude value is multiplied by the logarithmic result. Finally, the products corresponding to all frequency points are summed, and the negative of the sum is taken as the spectral entropy value. When a tool malfunctions, the complexity of the spectrum may change. Converting the time-domain signal to a frequency-domain signal visually displays the energy distribution of the signal at different frequencies. The spectral entropy value can reflect the complexity of the spectrum, providing frequency-domain characteristics for fault diagnosis. When a tool experiences wear or cracks, the energy distribution of the spectrum may change, and the spectral entropy value will change accordingly. By analyzing these changes, the health status of the tool can be determined.

[0036] The spindle current signal is synchronously averaged according to the spindle rotation cycle to obtain the instantaneous rate of change of the spindle current signal. Synchronous averaging means processing the spindle current signal according to the spindle rotation cycle, averaging the signal within multiple cycles to eliminate the influence of random noise and highlight the periodic changes related to spindle rotation. The instantaneous rate of change of the synchronously averaged signal is calculated using a first-order forward difference method. By calculating the instantaneous rate of change, the dynamic changes of the spindle during operation can be more accurately reflected, such as load changes and current fluctuations caused by faults, which helps to monitor the working status of the spindle and detect faults in a timely manner.

[0037] The eigenvectors include the envelope, short-time energy, peak factor, and spectral entropy value from the acoustic emission signal, the envelope, short-time energy, peak factor, and spectral entropy value from the vibration signal, and the instantaneous rate of change of the spindle current signal.

[0038] like Figure 2 The flowchart for tool status determination shown further illustrates that the alarm signal is triggered based on the abnormal score. The specific process is as follows:

[0039] The mean square error between the extracted feature vector and the reconstructed feature vector is defined as the anomaly score. By quantifying the difference between feature vectors through the mean square error, the deviation of the current state of the tool from the normal state (the normal feature pattern learned by the autoencoder) can be intuitively reflected. The larger the anomaly score, the more likely the tool state is to be abnormal, providing an effective quantitative basis for subsequent alarm judgment.

[0040] The reconstructed feature vector is the result of encoding and decoding the extracted feature vector using a one-dimensional convolutional autoencoder (DICO). A DICO is a neural network structure that combines the characteristics of one-dimensional convolution operations and autoencoders. An autoencoder consists of an encoder and a decoder. The encoder compresses the input data into a low-dimensional code, while the decoder reconstructs the code back into the original data space. One-dimensional convolution operations are suitable for processing one-dimensional data, such as time-series data. In tool condition monitoring, tool vibration signals can be considered as one-dimensional time-series data. A DICO can automatically learn important features from the data. Specifically, it can automatically learn the feature patterns of a tool under normal conditions. Through training on normal data, when a normal feature vector is input, the reconstructed feature vector is very close to the original feature vector with a small mean square error. However, when an abnormal feature vector is input, the reconstruction effect deteriorates, and the mean square error increases, enabling the autoencoder to effectively detect abnormal changes in tool condition.

[0041] A dynamic threshold is set based on the average and standard deviation of the abnormal scores within a preset time period. Specifically, the sum of the average of the abnormal scores and three times the standard deviation is set as the dynamic threshold. Since the normal state of the tool may fluctuate under different working stages and conditions, the dynamic threshold can be automatically adjusted according to the actual situation. Compared with the fixed threshold in the existing technology, the dynamic threshold can better adapt to the changes in the tool state, reduce false alarms and missed alarms, and improve the accuracy and reliability of automated production line management.

[0042] If the abnormal score exceeds the dynamic threshold, a tool status warning is issued, and spindle safety control is executed. This involves reducing the spindle feed rate and speed to preset safety values ​​and physically verifying the tool malfunction. Otherwise, the tool status is considered normal, and thermal drift compensation continues. The safety values ​​for the spindle feed rate and speed are preset by personnel based on experience, for example, they can be set to the minimum values ​​under normal spindle feed rate and speed conditions. Through spindle safety control, measures can be taken in the early stages of tool malfunction to prevent further deterioration of the malfunction, improve production safety and equipment lifespan, and ensure the stability of machining quality by continuing thermal drift compensation when the tool status is normal.

[0043] Further physical verification of tool failure follows the specific procedure below:

[0044] After the speed adjustment period has ended and the tool dimensions have stabilized, measuring the tool dimensions after the speed has stabilized can avoid measurement errors caused by speed instability, ensuring that the obtained tool dimension data is accurate and reliable, and providing a basis for subsequent judgment of the tool condition.

[0045] If the current tool size deviation is not greater than the tool size deviation limit, and the change in tool wear during this production process is not greater than the change in average tool wear, then the tool is considered to be in normal condition. The spindle state before the spindle safety control operation is restored, i.e., the previously set spindle feed rate and speed are restored, and thermal drift compensation continues. Tool size deviation represents the absolute value of the difference between the currently measured tool size and the average tool size, used to measure the degree of deviation between the current tool size and the average size under normal conditions. The change in tool wear is obtained by calculating the difference between the tool size before and after this production process. The average change in tool wear is the average value of the change in tool wear over a historical period. By setting tool size deviation limits and average tool wear changes as judgment criteria, the actual condition of the tool can be scientifically and objectively assessed. When the conditions are met, the tool is considered normal, allowing the machine tool to continue normal processing, ensuring production efficiency, while continuing thermal drift compensation effectively maintains machining accuracy and ensures stable product quality.

[0046] If the above conditions are not met, the tool condition is determined to be abnormal. An event log is recorded, including the timestamp of the abnormal tool condition, feature vector, tool dimensional deviation, and tool wear change, and a spindle emergency stop is triggered. A spindle emergency stop immediately halts the machine spindle's operation to prevent more serious machining accidents caused by tool failure, such as damage to the workpiece or machine tool components. Recording detailed event logs provides rich data support for technicians to analyze the causes of tool failures, helping to quickly locate problems, take targeted solutions, shorten troubleshooting time, and reduce production losses.

[0047] Furthermore, thermal drift compensation is performed at the edge nodes using the collected multi-source sensor data. The specific process is as follows:

[0048] During workpiece machining, the machine tool spindle and other components are affected by temperature changes, which can cause thermal deformation and lead to tool position displacement. Thermal drift compensation means that this displacement is corrected in real time to ensure machining accuracy.

[0049] Within a preset time period after workpiece production begins, a thermal drift compensation model is initialized based on the current ambient temperature and the CNC machine tool's standby status to predict thermal deformation trends. The model takes multiple temperature points (such as spindle temperature, ambient temperature, and coolant temperature) as input and outputs the thermal drift amount. Multiple linear regression modeling is used to describe the mapping between multiple temperature variables and thermal drift through linear relationships. This provides a basic prediction for subsequent thermal drift compensation, helps to anticipate potential thermal deformation in advance, and provides a reference direction for subsequent real-time compensation adjustments.

[0050] At the edge computing node, a Kalman filter is used to continuously update the parameters of the thermal drift compensation model based on real-time temperature data and spindle status data including spindle speed and load. This makes the model more closely match the actual operating state of the machine tool and outputs the real-time thermal drift compensation amount to the CNC system in a feedforward manner. The Kalman filter is an algorithm that performs optimal estimation of the machine tool state based on real-time temperature data and spindle status data. By updating the parameters of the drift compensation model with real-time data, the thermal drift compensation can dynamically adapt to changes in the machine tool operation process, improving the real-time performance and accuracy of the compensation, effectively reducing machining errors caused by thermal deformation, and improving machining accuracy.

[0051] like Figure 3 The diagram showing the physical information neural network architecture presents a hybrid prediction model architecture. Its input layer integrates features such as temperature, rotational speed, load, predicted drift, and actual drift using a multiple linear regression model. The LSTM hidden layer processes sequence information through multiple LSTM units, capturing long-term dependencies using mechanisms such as input gates, forget gates, output gates, and cell states. Each LSTM unit receives the hidden state h from the previous time step. t-1 and cell state Ct-1 and the current input X t Then output the hidden state h at the current moment. t and cell state C t Where σ is the sigmoid activation function used for gate control, and tanh is the hyperbolic tangent activation function used for generating candidate values; the output layer uses optimized Kalman filter parameters to further process the data; network training combines data loss and physical loss loss functions to optimize model parameters, thereby improving the accuracy and stability of predictions. On the cloud server, a physical information neural network based on a long short-term memory network is used to train the edge computing nodes on operating condition data, including temperature, spindle speed, and spindle load, as well as the predicted and actual thermal drift amounts from the thermal drift compensation model. The long short-term memory network can process time-series data and uncover patterns in data changes over time; the physical information neural network represents a neural network that integrates physical laws into the training process, enabling the network to follow physical laws while learning data features, thus improving the generalization ability and physical rationality of the thermal drift compensation model; through training on a large amount of historical and real-time data, the model can learn more complex and accurate thermal drift patterns, improving its ability to predict thermal drift under different operating conditions and providing a more reliable basis for subsequent optimization of the Kalman filter.

[0052] Key parameters are extracted from the trained physical information neural network model, including the state transition matrix, control input matrix, and process noise covariance of the Kalman filter. These key parameters are used to optimize the Kalman filter. The fully trained physical information neural network model contains rich thermal drift-related information. Key parameters related to the Kalman filter are extracted from the physical information neural network model. These parameters can reflect the dynamic characteristics and uncertainties in the machine tool thermal drift process. These parameters are used to optimize the Kalman filter, enabling it to more accurately describe the changes in the machine tool's thermal drift state and noise characteristics. The optimized Kalman filter can more accurately update the thermal drift compensation model parameters based on real-time data, further improving the accuracy and stability of thermal drift compensation and resulting in better compensation effects.

[0053] An adaptive optimizer is used to minimize the loss function of the Physical Information Neural Network (PIN). The PIN loss function is a weighted average of a data loss term and a physical loss term. The data loss term measures the difference between the predicted thermal drift and the actual thermal drift, represented by the compensation residual uploaded from edge computing nodes. The physical loss term includes the heat conduction equation and the thermal expansion equation, which are used to embed physical laws as constraints into the training process of the PIN, ensuring that the features learned by the PIN model conform to physical reality. By optimizing the loss function through the adaptive optimizer, the PIN model pursues data fitting accuracy while adhering to physical laws, improving its generalization ability and physical rationality, avoiding overfitting or non-physical discrepancies, and further enhancing its ability to predict and compensate for thermal drift. Model training and optimization are performed on cloud servers, while edge nodes handle real-time data processing and compensation, fully leveraging the advantages of edge computing and cloud computing to improve the efficiency and reliability of thermal drift compensation.

[0054] like Figure 4 The flowchart shown further illustrates how the compensation value for the cutting tool used in CNC machine tool production is obtained based on the workpiece dimension measurement data. The specific process is as follows:

[0055] After a single workpiece is produced, it is determined whether thermal drift exists during the monitoring process of the above production. The status of the machine tool during the processing is monitored and analyzed to see if there is thermal deformation of machine tool components due to temperature changes, which in turn causes a change in the relative position of the tool and the workpiece. Thermal drift phenomena are identified in a timely manner to provide a basis for taking different measurement methods in the future.

[0056] If thermal drift is present, it indicates a significant deviation in the workpiece dimensions. In this case, a full inspection of the workpiece dimensions is performed by a coordinate measuring machine (CMM). Otherwise, a random inspection of the workpiece dimensions is performed by a CMM. Full inspection can accurately obtain the dimensional data of the workpiece when thermal drift is present, providing a reliable basis for the subsequent accurate calculation of tool compensation values ​​and ensuring the machining accuracy of the workpiece. Random inspection, while ensuring a certain level of detection accuracy, reduces the workload of inspection, improves production efficiency, and lowers inspection costs.

[0057] Acquire initial measurement data of a preset number of workpiece dimensions, and obtain the average dimension based on the initial measurement data. Use this average dimension as the baseline parameter. The baseline parameter provides a benchmark for judging whether subsequent measurement data deviates. By comparing with the baseline parameter, the trend of workpiece dimension change can be seen intuitively. This helps to detect dimension deviations caused by factors such as tool wear and changes in machine tool condition in a timely manner, and provides an accurate basis for adjusting tool compensation values.

[0058] After obtaining the baseline parameters, a preset number of measurement data are obtained. If the preset number of measurement data shows a unidirectional offset trend relative to the baseline parameters, that is, all of the preset number of measurement data are greater than or less than the baseline parameters, it is determined that an offset has occurred; otherwise, it is determined that no offset has been sent. By comparing with the baseline parameters, it is possible to accurately determine whether there is a systematic deviation in the workpiece dimensions, avoiding misjudgment of dimensional offset due to data fluctuations caused by individual accidental factors. This improves the accuracy and reliability of the judgment and provides a scientific basis for whether the tool compensation value needs to be adjusted subsequently.

[0059] For measurement data that indicates a deviation, a new average dimension is obtained based on a preset number of measurement data. This new average dimension is then used as a new tool compensation value, and a tool compensation command is output to the CNC machine tool's control system. This updates the CNC machine tool's tool compensation parameters, enabling the CNC machine tool to correct the tool position according to the new compensation value during subsequent workpiece machining. This allows the machine tool to reflect the impact of machine tool machining status and tool wear on workpiece dimensions in real time and accurately, effectively improving the dimensional accuracy and consistency of CNC machine tool-machined workpieces.

[0060] like Figure 5 The diagram shows a structural schematic of an automated production line management system integrating intelligent manufacturing units. This embodiment of the invention provides an automated production line management system integrating intelligent manufacturing units, comprising:

[0061] The production process data acquisition module is used to synchronously collect multi-source sensor data reflecting the status of CNC machine tools during the workpiece production process, and to process the collected multi-source sensor data to extract feature vectors; thereby comprehensively and in real time obtaining the status information of CNC machine tools during the production process, providing an accurate and rich data foundation for subsequent analysis.

[0062] The feature vector analysis module is used to reconstruct the extracted feature vectors. Based on the extracted and reconstructed feature vectors, an anomaly score is obtained to quantify the degree of difference between the two, i.e., the extracted and reconstructed feature vectors. By reconstructing the feature vectors and comparing them with the original feature vectors, abnormal patterns in the data can be discovered, providing a quantitative basis for subsequent fault monitoring and helping to identify potential problems in a timely manner.

[0063] The machine tool fault monitoring module is used to trigger alarm signals based on anomaly scores, execute safety control operations on the spindle of the CNC machine tool, and perform physical verification of tool faults. If a tool fault is found, the spindle is triggered to stop immediately; otherwise, thermal drift compensation is performed at the edge nodes using collected multi-source sensor data to correct the tool path. It can respond promptly to abnormal conditions of the machine tool and ensure the safety of the machine tool and personnel through alarms and safety control operations. The thermal drift compensation function can improve machining accuracy even when the tool is fault-free but subject to thermal deformation.

[0064] The machine tool fault management module is used to obtain the compensation value of the tool used in the production of the workpiece on the CNC machine tool based on the measurement data of the workpiece size after the workpiece production is completed, and update the tool compensation parameters of the CNC machine tool; it adjusts the tool compensation parameters according to the actual machining results, so that the machine tool can more accurately control the tool position and machining size in subsequent machining, improve the consistency and stability of machining quality, and reduce machining errors caused by tool wear and other factors.

[0065] Example 2: The collected multi-source sensor data is processed to extract feature vectors. The extracted feature vectors are then reconstructed. Based on the extracted and reconstructed feature vectors, an anomaly score is obtained to quantify the degree of difference between the two (i.e., the extracted and reconstructed feature vectors). An alarm signal is triggered based on the anomaly score, and safety control operations are performed on the spindle of the CNC machine tool. The process also includes:

[0066] In precision production workshops with multiple varieties and small batches, the production status of each workpiece is different. It is difficult to collect a sufficient number of chipping samples for each working condition to train the model to identify tool chipping faults, or the prediction execution cycle is too short to continue using Example 1. Example 2 needs to be used for rapid analysis.

[0067] During a preset time period at the start of workpiece production, when the tool is in normal cutting mode, vibration and acoustic emission signals are collected during the normal cutting process. Based on the root mean square (RMS) of the vibration signal and the high-frequency count of the acoustic emission signal, vibration and acoustic emission thresholds are set respectively as benchmarks for subsequent judgment of whether the tool condition is abnormal. This accurately reflects the signal characteristics of the tool during normal operation and helps to accurately identify abnormal tool conditions. For vibration signals, the RMS of the vibration signal reflects the average energy level of the vibration signal and is used to determine whether the intensity of the vibration signal exceeds the normal range. For acoustic emission signals, the high-frequency count refers to the number of times the high-frequency components of the acoustic emission signal appear within a certain time. The appearance of high-frequency components is often related to certain abnormal conditions of the tool (such as chipping, accelerated wear, etc.), and the high-frequency count can help judge the tool condition. The vibration threshold is set as the sum of the average value of the RMS of the vibration signal and three times the standard deviation; the acoustic emission threshold is set as the sum of the average value of the high-frequency count of the acoustic emission signal and three times the standard deviation.

[0068] Real-time monitoring of vibration and acoustic emission signals. If the root mean square of the vibration signal exceeds the vibration threshold, or the high-frequency count of the acoustic emission signal exceeds the acoustic emission threshold, a tool status warning is issued. This allows for a rapid response when the tool exhibits initial abnormalities, preventing further deterioration of the fault. Otherwise, thermal drift compensation continues, ensuring the continuity and accuracy of the machining process.

[0069] After a tool status warning is issued, the corresponding load signal is collected and analyzed based on the currently executed process type. It is then determined whether the load signal meets the preset secondary confirmation conditions. If the secondary confirmation conditions for the corresponding process type are met, it indicates that the tool has indeed malfunctioned. Spindle safety control operations are then executed, and physical verification of the tool fault is performed. This physical verification is consistent with the physical verification of tool faults in Example 1. Otherwise, it indicates that the tool has not malfunctioned, and thermal drift compensation continues. Collecting and analyzing corresponding load signals for secondary confirmation based on different process types improves the accuracy of tool fault judgment because different processes have different load effects on the tool. Using targeted load signals and confirmation conditions can avoid misjudgments. Executing spindle safety control operations after confirming a tool fault effectively prevents the fault from escalating and causing greater damage to the machine tool and workpiece. Physical verification further confirms the fault situation, providing a basis for subsequent processing.

[0070] Specific load signals and secondary confirmation conditions are set for different processes, taking into full account the machining characteristics and tool load characteristics of each process. If the current process is a fine boring process, the spindle current signal is used as the load signal. This is because if the tool is abnormal during the fine boring process (such as chipping or wear), it will cause changes in the spindle load, which will be reflected in the current signal as an increase in the instantaneous change. The secondary confirmation condition for the fine boring process is that the instantaneous change of the spindle current signal is greater than the preset current change limit. The current change limit is set by the preset personnel based on experience, for example, it can be set as the average value of the current change limit of the fine boring process in the historical time period.

[0071] If the current operation type is milling the mating surface, the spindle current signal is taken as the load signal. Considering the special nature of the milling process, monitoring is only carried out within the angle window of the tool cutting into the workpiece. At this time, the cutting state of the tool and the workpiece is relatively stable and representative. The secondary confirmation condition for milling the mating surface is that the instantaneous change of the spindle current signal is greater than the preset current change limit. The current change limit is set by the preset personnel based on experience. For example, it can be set as the average value of the current change limit of milling the mating surface in the historical time period.

[0072] In the bolt hole drilling process, the torque feedback signal can more accurately reflect the load on the tool. If the current process type is bolt hole drilling, the torque feedback signal is taken as the load signal. The secondary confirmation condition for the bolt hole drilling process is that the instantaneous change of the torque feedback signal is greater than the preset torque change limit. The torque change limit is set by the preset personnel based on experience. For example, it can be set as the average value of the instantaneous change of the torque feedback signal of the bolt hole drilling process in the historical time period.

[0073] Example 3: Performing thermal drift compensation at edge nodes using acquired multi-source sensor data further includes:

[0074] The workshop environment is complex and contains interference factors such as vibrations from other equipment, personnel movement, and temperature changes, which may contaminate the data used to train the physical information neural network model in the cloud, leading to inaccurate model predictions. Therefore, Example 1 is no longer applicable, and Example 3 is required to perform thermal drift compensation.

[0075] Use a laser displacement sensor to measure the initial position or size of the workpiece or related machine tool components, and record it as the reference value;

[0076] During the production process, the temperature of the workpiece is measured using a temperature sensor at a preset first time interval, and the deformation of the workpiece relative to a reference value is measured using a laser displacement sensor. By acquiring temperature and deformation data in real time, data support is provided for subsequent analysis of the relationship between temperature and deformation, and the impact of temperature changes on workpiece deformation can be detected in a timely manner.

[0077] Based on the recorded temperature and deformation, the slope of the temperature-deformation linear function is obtained and updated using a linear regression algorithm. The temperature-deformation linear function is used to describe the relative relationship between the CNC machine tool and the thermal deformation. A quantitative relationship model between temperature and deformation is established through the linear regression algorithm. The update of the slope can reflect the changes in the relationship between temperature and deformation in real time, providing an accurate basis for subsequent prediction of deformation based on temperature changes.

[0078] If the slope fluctuation of the preset update number exceeds the slope fluctuation limit, it indicates that the relationship between temperature and deformation changes significantly at the current time interval, and the data may not be stable enough. The first time interval is extended to the second time interval. According to the second time interval, the current temperature and the corresponding deformation relative to the baseline value measured by the laser displacement sensor are recorded. Otherwise, the first time interval is continued, recording the current temperature and the corresponding deformation relative to the baseline value measured by the laser displacement sensor. The slope fluctuation range represents the standard deviation of the slope of the preset update number. The slope fluctuation limit is set by the preset personnel based on experience; for example, it can be set to the average value of the slope fluctuation range over a historical period. The first and second time intervals are set by the preset personnel based on experience; for example, the first time interval can be set to 5 minutes, and the second time interval to 10 minutes. The data recording time interval is dynamically adjusted according to the stability of the temperature-deformation relationship. When the slope fluctuation is large, the interval is extended to reduce the data recording frequency, avoiding the complexity of analysis due to excessive and unstable data. At the same time, it ensures that data can be obtained in a timely manner when the slope fluctuation is small, improving data quality and the accuracy of subsequent analysis.

[0079] The tool compensation value is obtained based on the updated deformation amount, and the tool compensation command is output to the CNC system to correct the tool path; the tool compensation value is equal to the negative of the deformation amount; the tool path is corrected in real time to effectively compensate for thermal deformation caused by temperature changes, ensure the machining accuracy and quality of the workpiece, and reduce the scrap rate caused by thermal drift.

[0080] Example 4: Obtaining the tool compensation value for workpiece production in a CNC machine tool based on workpiece dimension measurement data, further including:

[0081] In the practical example of waiting for a statistical event to occur, if the error source changes very rapidly, causing dimensional deviations within a few workpieces, then Example 1 is not applicable, and we need to switch to Example 4. Example 4 is predictive; by monitoring current and temperature in real time, it can immediately detect the trend of wear and thermal deformation after the first workpiece is produced, and provide compensation for the second workpiece.

[0082] The current fluctuation variance within a preset time period is obtained. Based on the wear coefficient and the current fluctuation variance deviation, the tool wear amount is obtained by multiplying them. The wear coefficient represents the influence coefficient of tool wear on workpiece size. The wear coefficient is determined experimentally and reflects the degree of influence of tool wear on workpiece size. The current fluctuation variance deviation represents the difference between the current fluctuation variance and the initial current fluctuation variance. The tool wear amount is obtained by obtaining the current fluctuation, which can directly measure the tool without stopping the machine, improving production efficiency and providing key data for subsequent tool compensation.

[0083] The current thermal drift of the spindle is obtained. The spindle thermal deformation value is obtained by multiplying the thermal drift value and the thermal deformation-size factor. The thermal deformation-size factor represents the influence coefficient of thermal deformation on the workpiece size. It is determined through experiments and reflects the degree of influence of spindle thermal deformation on workpiece size. The spindle thermal deformation value reflects the amount of influence of spindle thermal deformation on workpiece size. Accurately quantifying the influence of spindle thermal deformation on workpiece size provides data support on thermal deformation for tool compensation and further improves machining accuracy.

[0084] The compensation amount is obtained by performing deviation processing on the measured dimensional deviation and the predicted dimensional deviation. The measured dimensional deviation represents the deviation between the actual measured workpiece size and the theoretical size. In this application, deviation processing refers to difference calculation. The predicted dimensional deviation represents the sum of tool wear and spindle thermal deformation. The compensation amount represents the value of the tool compensation parameter that needs to be adjusted, reflecting the inaccuracy of tool size prediction, so as to eliminate the difference between the measured dimensional deviation and the predicted dimensional deviation. Taking into account the influence of factors such as tool wear and spindle thermal deformation on the workpiece size, the tool compensation parameter can be accurately adjusted by calculating the compensation amount, thereby improving the machining accuracy of individual workpieces.

[0085] The calculated compensation amount updates the tool compensation parameters for the next workpiece in the CNC system. This involves summing the compensation amount with the current tool compensation parameters to obtain new tool compensation parameters, and then outputting the tool compensation command to the CNC machine tool's control system. When machining the next workpiece, the CNC system controls the tool movement according to the new tool compensation parameters, thereby compensating for factors such as tool wear and spindle thermal deformation. This achieves dynamic updating of the tool compensation parameters, ensuring that each workpiece can be machined based on the current tool condition and machine tool thermal deformation, thus guaranteeing the stability of machining quality.

[0086] After the current batch of workpieces is completed, the measured dimensional deviation and the predicted dimensional deviation of the batch of workpieces are obtained. Using the least squares method, based on the error between the measured and predicted dimensional deviations of the batch of workpieces, the wear coefficient and the thermal deformation-size coefficient are optimized and updated to minimize the sum of squares of the errors between the measured and predicted dimensional deviations. By continuously adjusting these two coefficients, the predicted dimensional deviation is made closer to the measured dimensional deviation, improving the accuracy of subsequent predictions. As production progresses, the coefficients are continuously optimized to make tool compensation more adaptable to the actual machining environment and conditions such as tool wear and machine tool thermal deformation, further improving machining accuracy and the stability of the production process.

[0087] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0088] It should be understood that determining B based on A does not mean determining B solely based on A; it also means determining B based on A and / or other information.

[0089] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0090] The above description is only an optional embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for managing an automated production line integrating intelligent manufacturing units, characterized in that, The method includes: S1. During the workpiece production process, multi-source sensor data reflecting the status of CNC machine tools are collected, and the collected multi-source sensor data is processed to extract feature vectors. S2, reconstruct the extracted feature vector, and obtain the anomaly score based on the extracted feature vector and the reconstructed feature vector to quantify the degree of difference between the two; S3 triggers an alarm signal based on an abnormal score, performs safety control operations on the spindle in the CNC machine tool, and performs physical verification of tool failure. If a tool failure is detected, the spindle is stopped immediately; otherwise, thermal drift compensation is performed at the edge nodes using the collected multi-source sensor data to correct the tool path. S4. After the workpiece is produced, obtain the compensation value of the tool used in the CNC machine tool for workpiece production based on the measurement data of the workpiece size, and update the tool compensation parameters of the CNC machine tool. The compensation value of the cutting tool used in CNC machine tool production is obtained based on the measurement data of the workpiece dimensions. The specific process is as follows: After the production of a single workpiece is completed, determine whether thermal drift exists during the monitoring process of the above production. If thermal drift exists, the coordinate measuring machine performs a full inspection of the workpiece dimensions; otherwise, the coordinate measuring machine performs a random inspection of the workpiece dimensions. Acquire initial measurement data of a preset number of workpiece dimensions, and obtain the average dimension value based on the initial measurement data, using the average dimension value as the baseline parameter; After obtaining the baseline parameters, a preset number of measurement data is obtained. If the preset number of measurement data shows a unidirectional offset trend relative to the baseline parameters, that is, the preset number of measurement data are all greater than the baseline parameters or all less than the baseline parameters, it is determined that an offset has occurred; otherwise, it is determined that no offset has been sent. For measurement data that indicates a deviation, a new average dimension is obtained based on a preset number of measurement data, and this new average dimension is used as a new compensation value to update the tool compensation parameters of the CNC machine tool.

2. The automated production line management method integrating intelligent manufacturing units as described in claim 1, characterized in that, The specific process for processing the collected multi-source sensor data to extract feature vectors is as follows: The acoustic emission signal in the multi-source sensor data is subjected to a first bandpass filter, and the vibration signal in the multi-source sensor data is subjected to a second bandpass filter to filter out low-frequency mechanical vibration and high-frequency noise. The pass range of the first bandpass filter is greater than the pass range of the second bandpass filter; Hilbert transforms were performed on the bandpass filtered acoustic emission signal and vibration signal respectively to extract the signal envelope and capture the periodic impact characteristics caused by tool chipping. Wavelet transform is used to decompose the bandpass filtered signal and extract the short-time energy and peak factor within a preset time window. Fast Fourier Transform is performed on the bandpass filtered acoustic emission signal and vibration signal to obtain the spectrum, and the spectral entropy value of the spectrum is obtained. The spindle current signal is synchronously averaged according to the spindle rotation cycle to obtain the instantaneous rate of change of the spindle current signal; The feature vector includes the envelope, short-time energy, peak factor, and spectral entropy value from the acoustic emission signal, the envelope, short-time energy, peak factor, and spectral entropy value from the vibration signal, and the instantaneous rate of change of the spindle current signal.

3. The automated production line management method integrating intelligent manufacturing units as described in claim 1, characterized in that, The specific process for triggering the alarm signal based on the abnormal score is as follows: The mean squared error between the extracted feature vector and the reconstructed feature vector is defined as the outlier score. The reconstructed feature vector is the result of reconstructing the extracted feature vector after encoding and decoding by a one-dimensional convolutional autoencoder. A dynamic threshold is set based on the average and standard deviation of abnormal scores within a preset time period; If the abnormal score is greater than the dynamic threshold, a tool status warning is issued, and a spindle safety control operation is executed, which means reducing the spindle feed rate and speed to preset safety values ​​and performing physical verification of tool failure; otherwise, thermal drift compensation continues.

4. The automated production line management method integrating intelligent manufacturing units as described in claim 3, characterized in that, The physical verification process for the tool failure is as follows: After the spindle speed adjustment stabilization period ends, measure the tool size. If the current tool size deviation is not greater than the tool size deviation limit, and the change in tool wear during this production process is not greater than the change in average tool wear, then the tool condition is determined to be normal. Restore the spindle condition before the spindle safety control operation and continue thermal drift compensation. If the current tool size deviation is greater than the tool size deviation limit, or if the change in tool wear during this production process is greater than the average change in tool wear, then the tool condition is determined to be abnormal. An event log is recorded, including the timestamp of the time of the abnormal tool condition, feature vector, tool size deviation, and change in tool wear, and the spindle emergency stop is triggered.

5. The automated production line management method for integrated intelligent manufacturing units as described in claim 1, characterized in that, The specific process for performing thermal drift compensation at edge nodes using collected multi-source sensor data is as follows: Within a preset time period after the start of workpiece production, the thermal drift compensation model is initialized based on the current ambient temperature and the standby status of the CNC machine tool to predict the trend of thermal deformation. At the edge computing node, a Kalman filter is used to update the parameters of the thermal drift compensation model based on the real-time collected temperature data and spindle status data, and the real-time thermal drift compensation amount is output to the CNC system in a feedforward manner. On the cloud server, a physical information neural network based on a long short-term memory network is used to train the edge computing node with operating data including temperature, spindle speed and spindle load, thermal drift amount predicted by the thermal drift compensation model and the actual thermal drift amount. Key parameters are extracted from the trained physical information neural network model, including the state transition matrix of the Kalman filter, the control input matrix and the process noise covariance. The loss function of the physical information neural network is minimized using an adaptive optimizer. The loss function of the physical information neural network is composed of a weighted sum of data loss terms and physical loss terms. The data loss term measures the difference between the thermal drift predicted by the thermal drift compensation model and the actual thermal drift, which is represented by the compensation residual uploaded by the edge computing node. The physical loss term is used to embed physical laws as constraints into the training process of the neural network.

6. The automated production line management method integrating intelligent manufacturing units as described in claim 1, characterized in that, The alarm signal triggered based on the abnormal score is used to execute the safety control operation of the spindle in the CNC machine tool, which then includes: Vibration signals and acoustic emission signals during the cutting process are collected within a preset time period after the start of workpiece production. Based on the root mean square of the vibration signal and the high-frequency count of the acoustic emission signal, vibration threshold and acoustic emission threshold are set respectively. Real-time monitoring of vibration and acoustic emission signals. If the root mean square of the vibration signal exceeds the vibration threshold, or the high-frequency count of the acoustic emission signal exceeds the acoustic emission threshold, a tool status warning is issued; otherwise, thermal drift compensation continues. After determining the tool status warning, the corresponding load signal is collected and analyzed according to the current process type. If the secondary confirmation condition of the corresponding process type is met, the spindle safety control operation is executed and the physical verification of the tool fault is performed. Otherwise, thermal drift compensation continues. If the current process type is fine boring, the spindle current signal is taken as the load signal. The secondary confirmation condition of the fine boring process is that the instantaneous change of the spindle current signal is greater than the preset current change limit. If the current process type is milling the mating surface, the spindle current signal is taken as the load signal, and monitoring is only performed within the angle window of the tool cutting into the workpiece. The secondary confirmation condition for the milling mating surface process is that the instantaneous change of the spindle current signal is greater than the preset current change limit. If the currently executed process is the drilling of bolt holes, then the torque feedback signal is taken as the load signal. The secondary confirmation condition for the drilling of bolt holes is that the instantaneous change in the torque feedback signal is greater than the preset torque change limit.

7. The automated production line management method integrating intelligent manufacturing units as described in claim 4, characterized in that, The step of performing thermal drift compensation at edge nodes using acquired multi-source sensor data also includes: Acquire the values ​​recorded by the laser displacement sensor and record them as the reference values; During the workpiece production process, the current temperature and the corresponding deformation relative to the reference value, measured by the laser displacement sensor, are recorded at a preset first time interval. Based on the recorded temperature and deformation, the slope of the temperature-deformation linear function is obtained and updated through a linear regression algorithm. The temperature-deformation linear function is used to describe the relative relationship between the CNC machine tool and the thermal deformation. If the fluctuation range of the slope of the preset update number is greater than the slope fluctuation limit, the first time interval is extended to the second time interval. According to the second time interval, the current temperature and the corresponding deformation relative to the reference value measured by the laser displacement sensor are recorded. Otherwise, the current temperature and the corresponding deformation relative to the reference value measured by the laser displacement sensor are recorded according to the first time interval. The tool compensation value is obtained based on the updated deformation amount, and the tool compensation command is output to the CNC system to correct the tool path.

8. The automated production line management method integrating intelligent manufacturing units as described in claim 1, characterized in that, The method of obtaining the tool compensation value for workpiece production in a CNC machine tool based on workpiece size measurement data also includes: The current fluctuation variance within a preset time period is obtained. Based on the wear coefficient and the current fluctuation variance deviation, the tool wear amount is calculated. The wear coefficient represents the influence coefficient of tool wear on workpiece size, and the current fluctuation variance deviation represents the difference between the current fluctuation variance and the initial current fluctuation variance. Obtain the current thermal drift of the spindle, and calculate the thermal deformation value of the spindle based on the thermal drift and the thermal deformation-size factor, wherein the thermal deformation-size factor represents the influence coefficient of thermal deformation on the workpiece size. The compensation amount is obtained by performing deviation processing on the measured size deviation and the predicted size deviation; The calculated compensation amount is used to update the tool compensation parameters for the next workpiece in the CNC system. After the current batch of workpieces is produced, the wear coefficient and thermal deformation-size coefficient are optimized and updated based on the error between the measured dimensional deviation and the predicted dimensional deviation of the produced workpieces using the least squares method.

9. An automated production line management system integrating intelligent manufacturing units, used to implement the automated production line management method integrating intelligent manufacturing units as described in any one of claims 1-8, characterized in that, The system includes: The production process data acquisition module is used to collect multi-source sensor data reflecting the status of CNC machine tools during the workpiece production process, and to process the collected multi-source sensor data to extract feature vectors. The feature vector analysis module is used to reconstruct the extracted feature vectors, and based on the extracted feature vectors and the reconstructed feature vectors, an anomaly score is obtained to quantify the degree of difference between the two. The machine tool fault monitoring module is used to trigger alarm signals based on abnormal scores, execute safety control operations of the spindle in CNC machine tools, and perform physical verification of tool faults. If a tool fault is found, the spindle is triggered to stop immediately; otherwise, thermal drift compensation is performed at the edge nodes using the collected multi-source sensor data to correct the tool path. The machine tool fault management module is used to obtain the compensation value of the tool used in the production of the workpiece in the CNC machine tool based on the measurement data of the workpiece size after the workpiece production is completed, and to update the tool compensation parameters of the CNC machine tool.

Images