Flow anomaly identification method for numerical control machine tool protocol

Abstract

The invention relates to the technical field of flow anomaly identification, and discloses a flow anomaly identification method for a numerical control machine tool protocol, which comprises the following steps of: 1, multi-source data acquisition: acquiring various types of data from a numerical control machine tool and a control system thereof by arranging a data acquisition module, the data sources include network protocol flow data, machine tool sensor data, operation log data and equipment environment data, and the data formats of the data sources are kept uniform; the collected original data is subjected to standardization processing, the problems of data dimension inconsistency, noise interference and missing values are solved, and alignment of the data on a time axis is guaranteed through a synchronization algorithm; through a multi-source data acquisition and high-precision sensor system, the real-time monitoring of the flow of each key fluid system of the numerical control machine tool is realized. And by combining a plurality of feature extraction methods, the traffic anomaly can be effectively identified.

Description

Technical Field

[0001] The present invention relates to the technical field of flow anomaly recognition, and in particular to a flow anomaly recognition method for a numerical control machine tool protocol. Background Art

[0002] With the development of CNC technology, CNC machine tools have been widely used in modern manufacturing. With their high precision, high efficiency and degree of automation, CNC machine tools are widely used in aviation, automobiles, molds, precision instruments and other fields, becoming important equipment in production. However, with the increase of machine tool working time and the complex changes in working conditions, machine tools will face various potential problems in long-term operation. In particular, if the flow of machine tool fluid systems (such as hydraulic systems, coolant systems, etc.) is abnormal, if it is not discovered and handled in time, it may lead to equipment failure, reduced processing accuracy, and even cause more serious mechanical damage, affecting production efficiency and quality.

[0003] At present, the flow monitoring of CNC machine tools mainly relies on simple sensor data acquisition and alarm systems. However, due to the large differences in the operating characteristics of CNC machine tools under different working conditions, traditional flow anomaly detection methods are usually based on fixed thresholds for monitoring. Although these methods can work in some cases, they cannot fully consider the changes in the working environment of the machine tools and various complex working conditions, so their detection accuracy is limited and they are prone to false alarms and missed alarms. In addition, existing technologies often only focus on simple monitoring of flow, without in-depth analysis and fault location, and lack in-depth analysis and intelligent processing of flow data. Therefore, a single flow threshold judgment cannot cope with the complex changes of CNC machine tools under different working conditions.

[0004] The existing technology has significant problems in flow anomaly detection: since most detection systems use a monitoring method based on fixed thresholds, when the working environment of the machine tool changes or the operating conditions are different from the preset, false alarms or missed alarms may occur, and potential sources of faults may not be discovered in time. In addition, the existing technology often lacks joint analysis of multi-source data, and fails to effectively use advanced algorithms such as machine learning to intelligently analyze and process flow data, which greatly reduces the accuracy and timeliness of flow anomaly detection. To address this problem, a flow anomaly identification method based on multi-source data collection, machine learning algorithms and real-time monitoring is proposed, aiming to solve the problems of false alarms and missed alarms in flow anomaly detection through more accurate data analysis, and to accurately locate faults, thereby improving the fault warning capability and production efficiency of CNC machine tools. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides a flow anomaly identification method for CNC machine tool protocols, which solves the problems of false alarms and missed alarms in traditional CNC machine tool flow anomaly detection methods.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for identifying flow anomalies for CNC machine tool protocols, comprising the following steps: Step 1: Multi-source data collection: By setting up a data collection module, various types of data are collected from CNC machine tools and their control systems, including network protocol flow data, machine tool sensor data, operation log data, and equipment environment data, and the data format of each data source remains unified; Step 2: Data preprocessing and standardization: Standardize the collected raw data to solve the problems of inconsistent data dimensions, noise interference and missing values, and ensure the alignment of data on the time axis through synchronization algorithms; Step 3: Data feature extraction and representation: Perform feature engineering processing based on the collected data, extract the time series features, frequency domain features and statistical features in the network traffic data, extract the working status features in the sensor data, aggregate all the features to form a high-dimensional feature vector; Step 4: Traffic anomaly detection: The feature vector is input through the deep neural network model, and the spatial features of the data are extracted using a multi-layer convolutional neural network. The long short-term memory network is combined to capture the time-dependent characteristics, and anomaly recognition is performed using an adaptive threshold algorithm.

[0007] Preferably, the convolutional neural network in the deep neural network model is used to extract local spatial features of input feature data, and the long short-term memory network is used to extract time series features of the data, so as to accurately model multi-dimensional and multi-time series data.

[0008] Preferably, the data feature extraction step further includes performing wavelet transform on the network traffic data to convert it into time-frequency features, and performing Fourier transform and sliding average processing on the sensor data to extract frequency domain features and volatility features.

[0009] Preferably, the data preprocessing step also includes denoising the collected data, using wavelet packet transform to remove high-frequency noise, and using Kalman filtering algorithm to smooth abnormal fluctuations in the data.

[0010] Preferably, the flow anomaly detection step further includes using an adaptive threshold method to establish a flow fluctuation model through historical data, and dynamically adjust the anomaly detection threshold by comparing the model output with real-time data.

[0011] Preferably, the flow anomaly detection step also includes using a reinforcement learning algorithm to perform online optimization on the anomaly detection model and dynamically adjust model parameters to adapt to flow change characteristics of different machine tools under different loads and operating conditions.

[0012] Preferably, the flow anomaly detection step further includes clustering analysis based on time series data, using a K-means clustering algorithm to classify the flow data, and determining whether an anomaly occurs based on the classification results.

[0013] Preferably, the method further comprises monitoring the status of the machine tool in real time, triggering an alarm mechanism when a flow anomaly is detected, and outputting detailed fault location information according to different anomaly types.

[0014] Preferably, the method can model the fluctuation pattern of network traffic in the CNC machine tool protocol, identify abnormal changes in network communication indicators such as data packet size, packet loss rate, delay and bandwidth, and perform correlation analysis with the operating status of the machine tool.

[0015] The present invention provides a method for identifying flow anomalies for CNC machine tool protocols. It has the following beneficial effects: 1. The present invention realizes real-time monitoring of the flow of each key fluid system of a CNC machine tool through multi-source data acquisition and a high-precision sensor system. Combined with a variety of feature extraction methods, it can effectively identify flow anomalies. This multi-dimensional anomaly identification method greatly improves the accuracy of flow anomaly detection, ensures that potential problems can be discovered in a timely manner, and avoids missed detections or false alarms that may be caused by traditional methods.

[0016] 2. The present invention can not only detect abnormal changes in flow, but also combine multi-source data to perform fault diagnosis and location through machine learning algorithms. Based on the joint analysis of multiple signals such as flow, pressure, and temperature, the system can accurately locate the source of the problem and quickly identify the specific location where the equipment failure occurred. This multi-level analysis and location method helps to reduce the time and cost of fault handling and improve equipment maintenance efficiency.

[0017] 3. The system of the present invention adopts the design of online monitoring and model adaptive optimization. By continuously collecting and analyzing real-time data, the system can automatically adjust the model parameters according to the changes in the equipment working conditions, ensuring that the efficient flow anomaly detection function can be maintained under different working environments. This adaptive optimization mechanism not only improves the stability of the system under complex working conditions, but also can update the model at any time according to new data, thereby improving the long-term operation reliability of the system.

[0018] 4. The present invention can timely identify abnormalities before a CNC machine tool fails through early warning and rapid location of abnormal flow, thus avoiding sudden equipment shutdown. Maintenance personnel on the production line can quickly take effective repair measures based on the fault location information provided by the system, thus reducing equipment downtime and ensuring efficient operation of the production line, thereby effectively reducing production losses caused by equipment failure.

[0019] 5. The present invention combines big data analysis, machine learning and intelligent algorithms to conduct in-depth analysis of the flow data of CNC machine tools, and can dynamically adjust the operation strategy of the equipment and optimize the production process. The system can not only identify and handle anomalies, but also provide decision support for production managers, such as optimizing the flow parameters, thereby further improving the working efficiency and production quality of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] Please refer to the attached Figure 1 , an embodiment of the present invention provides a method for identifying flow anomalies for a CNC machine tool protocol, comprising: 1. Multi-source data collection In order to realize the flow anomaly identification of CNC machine tool protocols, it is necessary to first collect data from multiple data sources. These data sources include the working status data of CNC machine tools, sensor data, network flow data, equipment operation logs, etc., covering all kinds of information during the operation of the machine tools, which can provide the necessary basis for subsequent anomaly identification. Through the fusion of multi-source data, the operating status of the machine tools can be comprehensively monitored and analyzed from different dimensions, thereby improving the accuracy and reliability of flow anomaly detection.

[0023] In this embodiment, the data acquisition module acquires various types of data from the CNC machine tool and its control system in real time by setting up various sensors and data acquisition devices, including but not limited to network protocol flow data, equipment sensor data, machine tool operation logs, environmental data, etc. The information provided by each data source complements each other and provides comprehensive support for traffic anomaly identification. Specifically, it includes the following types of data sources: 1.1 Network traffic data Network traffic data refers to the network communication data between the machine tool and the external control system, remote monitoring system and other equipment. The collection of network traffic data can reflect the changes in the communication protocol and potential abnormalities at the network level during the operation of the machine tool. Network traffic data includes but is not limited to the following: Packet size: The number of bytes in each packet, which is an important indicator to measure traffic fluctuations.

[0024] Transmission delay: The time it takes for a data packet to reach its destination from the source. Long delays may indicate a network bottleneck or transmission interruption.

[0025] Packet loss rate: The ratio of data packets lost during data transmission to the total number of data packets. An abnormal packet loss rate may indicate a network failure or device communication problem.

[0026] Bandwidth usage: reflects the usage of network bandwidth. When the bandwidth is too high or too low, it may cause abnormal traffic.

[0027] In this embodiment, network traffic data is collected by setting up a special network traffic monitoring module, which can record every data interaction between the machine tool and the control system in real time. Through the collected network traffic data, traffic fluctuation patterns can be identified and basic data can be provided for the anomaly detection model.

[0028] 1.2 Device sensor data Equipment sensor data includes physical parameter data generated by the machine tool during operation, such as temperature, pressure, vibration, speed, load, etc. These data reflect the working status of the machine tool and the load of the equipment. The collection of sensor data helps to determine the stability of the machine tool operation and whether there are abnormal conditions such as overload and overheating. Specific sensor data includes: Temperature: The temperature of various parts of the machine tool changes. Excessive temperature may cause equipment performance degradation or failure.

[0029] Pressure: Pressure data of the machine tool hydraulic system or pneumatic system. Abnormal pressure fluctuations may indicate equipment failure.

[0030] Vibration: The vibration data of the machine tool while it is running. Abnormal vibration patterns may indicate mechanical failure or imbalance.

[0031] Speed ​​and load: The speed and load data of the machine tool spindle or feed system. Speed ​​and load that deviate from the normal range may indicate system failure.

[0032] In this embodiment, the equipment sensor data is collected in real time through the sensor module, and the sensor module can directly interface with the machine tool control system to monitor the operating status of the machine tool in real time. Through the collection of sensor data, the operating environment and load conditions of the machine tool can be monitored in real time to provide support for flow anomaly detection.

[0033] 1.3 Operation log data Operation log data is the operation history information recorded by the machine tool control system, including operator instructions, machine tool startup and shutdown time, fault alarm information, etc. These log data reflect the key events in the machine tool operation process and provide background information for abnormal flow detection. The specific contents include: Operation commands: various operation instructions issued by operators, such as start, pause, stop, parameter setting, etc.

[0034] Fault alarm: Alarm information triggered when a machine tool fails, which may include system failure, hardware failure or software abnormality.

[0035] State switching record: a record of changes in machine tool status, such as when the machine tool switches from idle state to processing state, or from processing state to maintenance state.

[0036] In this embodiment, the operation log data is collected by the log recording module in the control system, and the log information can be stored in the database for subsequent analysis. By analyzing the operation log, the operation background can be provided for the flow anomaly detection, helping to determine the potential cause of the abnormal flow.

[0037] 1.4 Environmental data Environmental data refers to the physical parameters that affect the working environment of the machine tool, such as temperature, humidity, air pressure, etc. These environmental factors may sometimes affect the normal operation of the machine tool and cause abnormal flow fluctuations. Specifically include: Temperature and humidity data: Changes in ambient temperature and humidity may affect the electronic components of machine tool equipment and thus affect network communications.

[0038] Air pressure data: Changes in air pressure may affect the working status of the hydraulic system or pneumatic system, and thus affect the operating status of the machine tool.

[0039] In this embodiment, environmental data is collected by setting up environmental monitoring sensors, which can collect parameters such as temperature, humidity and air pressure in the working area of ​​the machine tool in real time. By collecting environmental data, the flow anomaly detection model can be further optimized to consider the interference of environmental factors.

[0040] 1.5 Implementation of data acquisition module In order to ensure the accuracy and real-time performance of data collection, the data collection module in this embodiment adopts a distributed architecture design. Each data source has an independent data collection node, and all nodes are connected to the central processing unit through a unified interface. The data collection module includes the following parts: Data collection node: Each data source collects data through dedicated sensors or network monitoring equipment, and transmits the collected data to the central processing unit. Each node can work independently to ensure the real-time and reliability of the data collection process.

[0041] Data transmission protocol: Data is transmitted between the data acquisition node and the central processing unit through a standardized network protocol. Commonly used protocols include TCP / IP protocol, Modbus, OPC, etc., to ensure efficient data transmission and compatibility.

[0042] Data storage and processing: The collected raw data will be stored in the data storage system, which can be a relational database or a distributed database. The data processing module pre-processes the stored data, such as denoising, standardization, synchronization, etc., to provide clean input data for subsequent analysis and modeling.

[0043] 1.6 Data Synchronization and Fusion Since the acquisition time of each data source may have different timestamps or different time resolutions of the data, it is necessary to time align and fuse all data sources. Specifically: Time alignment: Align the timestamps from different data sources to ensure that the timestamps of the data are consistent. If there is data missing, fill in the missing data through interpolation or interpolation methods.

[0044] Data fusion: Through feature fusion technology, multi-source data is merged into a high-dimensional feature vector to ensure that the subsequent anomaly detection algorithm can process comprehensive information from multiple data sources. Data fusion can be done by simple splicing or by processing through dimensionality reduction algorithms (such as PCA). In this embodiment, through the multi-source data collection module, data from multiple sources such as network traffic, device sensors, operation logs, and environmental data can be collected in real time. The fusion of various types of data can not only improve the accuracy of anomaly identification, but also provide more comprehensive information support to better detect and identify traffic anomalies. Steps such as data preprocessing, synchronization, and feature extraction ensure that the data can be effectively applied in subsequent analysis, thereby laying the foundation for the accurate identification of traffic anomalies.

[0045] 2. Data preprocessing and standardization After multi-source data collection, the next step is to preprocess and standardize the collected data. Due to the diverse data sources and different data types (such as network traffic data, sensor data, operation logs, etc.), these data may differ in time scale, unit, and distribution. In order to ensure the uniformity of the data and eliminate these differences, the raw data needs to be cleaned, denoised, aligned, and standardized through data preprocessing. Data standardization can not only eliminate dimensional differences, but also improve the stability and accuracy of subsequent feature extraction and anomaly detection models.

[0046] In this embodiment, the data preprocessing and standardization process mainly includes the following steps: data synchronization and alignment, noise removal, missing value processing and standardization processing. These steps can ensure the consistency of data when it is input into the model, thereby improving the effect of anomaly detection.

[0047] 2.1 Data synchronization and alignment In the process of multi-source data collection, data from different sensors, control systems and devices have different timestamps and collection frequencies. In order to ensure the comparability of data, data from different sources need to be aligned to a unified time scale. The process of data synchronization and alignment first includes: Time alignment: Adjust the timestamps of all data sources to a unified time base through interpolation or resampling methods. The purpose of time alignment is to make events from different data sources correspond to the same time point to avoid errors caused by time inconsistency. For example, for network traffic data and sensor data, linear interpolation can be used to upsample low-frequency data to the same time resolution as high-frequency data.

[0048] Interpolation formula: ; in, is the estimated value at the current time point, and are known values ​​at the previous and next time points, and Timestamps of previous and next time points.

[0049] Data time window alignment: For sensor data collected at a high frequency and log data collected at a low frequency, fixed time windows are used for alignment. That is, the data is divided into fixed time windows for processing to ensure that the data in each window is comparable.

[0050] 2.2 Noise Removal Due to external environmental factors, sensor errors, network fluctuations, etc., the raw data may contain noise. In the data preprocessing stage, denoising technology is needed to smooth the data to eliminate the interference of abnormal fluctuations on the anomaly detection model. Common denoising methods include: Kalman filter: Kalman filter is a recursive optimal estimation method suitable for removing noise from sensor data. Its basic idea is to use weighted average to obtain a more accurate estimate based on prior information and new observations.

[0051] The update equation of the Kalman filter is: ; in, is an estimate of the current state, is the Kalman gain, is the new observation.

[0052] Wavelet transform: Wavelet transform can effectively extract low-frequency information from time domain data and remove high-frequency noise by analyzing the frequency components of the data. This method is particularly suitable for denoising vibration data and time series data.

[0053] The formula for wavelet transform is: ; Among them, ψ(t) is the wavelet function, s is the scale factor, and τ is the translation factor. By selecting an appropriate wavelet function (such as Morlet wavelet or Haar wavelet), filtering of data in different frequency bands can be achieved.

[0054] 2.3 Missing Value Processing In the actual data collection process, some data may be missing due to hardware failure, network problems, etc. In order to avoid the impact of missing data on subsequent analysis, missing values ​​need to be processed. Common missing value processing methods include: Interpolation: For time series data, interpolation can be used to fill in missing values. For example, for continuous sensor data, linear interpolation or spline interpolation can be used to estimate missing values.

[0055] The linear interpolation formula is: ; in, is the estimated value of the missing points, and are the data values ​​at adjacent time points, is the timestamp of the missing value, and are the timestamps of adjacent data points.

[0056] Deletion method: For data with a large proportion of missing data or a large impact, you can directly delete the data row or column. The deletion method is suitable for situations where the data loss is small and does not affect the overall analysis.

[0057] 2.4 Standardization Since data from different data sources have different units and dimensions, standardization is required to enable them to be compared and integrated at the same scale. Standardization can eliminate the impact of different dimensions and ensure that subsequent feature extraction and analysis models can better process these data.

[0058] Z-score standardization: A common standardization method is Z-score standardization, which maps the data to a standard normal distribution with zero mean and unit standard deviation by adjusting the mean and standard deviation of the data. The standardization formula is: ; Among them, x is the original data, μ is the mean of the data, and σ is the standard deviation of the data. Through Z-score standardization, the distribution of the data will be adjusted to zero mean and unit standard deviation, thereby eliminating the scale differences between different data sources.

[0059] Min-Max Normalization: Another common normalization method is Min-Max Normalization, which scales the data to a fixed range (such as [0,1]), which is suitable for situations where there are strict requirements on the input range of the model. The normalization formula is: ; in, and They are the minimum and maximum values ​​of the data. Through Min-Max normalization, the data can be mapped to a uniform numerical range.

[0060] 2.5 Data deduplication and consistency processing During the data collection and transmission process, duplicate data or inconsistent data records may appear. In order to ensure the accuracy and integrity of the data, these duplicate data need to be deduplicated to ensure that the data at each time point in the data is unique and consistent.

[0061] Deduplication: Remove duplicate records by checking timestamps and data content. For example, if data at the same time point is recorded multiple times, the most accurate data record is retained.

[0062] Consistency check: Check the consistency of data to ensure that data from different data sources in the same time period are consistent. For example, if there is a conflict between sensor data and network traffic data at the same time point, it should be corrected or marked as anomaly through models or rules.

[0063] 3. Data feature extraction and representation After data preprocessing and standardization, the next step is to extract meaningful features from the processed data. The purpose of data feature extraction is to mine key patterns and information from raw data that are helpful for traffic anomaly identification. This process can convert high-dimensional, multimodal data into low-dimensional, discriminative features, thereby providing effective input for subsequent anomaly detection algorithms. Since the goal of the present invention is to identify traffic anomalies based on CNC machine tool protocols, special attention should be paid to the machine tool operating status, changes in communication protocols, and related timing characteristics during the feature extraction stage.

[0064] In this embodiment, data feature extraction and representation mainly include time series feature extraction, frequency domain feature extraction and multimodal data fusion. Through these methods, key features reflecting traffic anomalies can be extracted from different data sources, providing effective support for traffic anomaly identification models.

[0065] 3.1 Temporal feature extraction Time series data refers to data that changes over time, including network traffic, device sensor data, operation logs, etc. The purpose of time series feature extraction is to extract dynamic features that reflect traffic anomalies from time series data. These features can help identify fluctuations in traffic patterns and can reflect changes in the operating status of machine tools to a certain extent.

[0066] In this embodiment, the time series feature extraction mainly includes the following steps: Trend analysis: By analyzing the trend of time series data, we can identify whether the data shows a long-term upward or downward trend. Common trend analysis methods include sliding average method, least squares method, etc. Trend analysis can help identify whether there are abnormal load fluctuations or long-term deviations in the working mode during machine tool operation.

[0067] The calculation formula for the sliding average is: ; in, is a sliding average of length n, It's time point The original data value of i.

[0068] Periodicity analysis: Periodicity analysis can reveal whether there are periodic changes in the data. For some machine tools that operate periodically, traffic fluctuations may have obvious periodic characteristics. For example, the load changes of the equipment or network traffic may fluctuate regularly according to the working cycle of the machine tool. Using Fourier transform or wavelet transform can effectively extract periodic features.

[0069] Volatility analysis: Volatility analysis reveals the degree of fluctuation of traffic data by calculating statistics such as standard deviation and variance of data. For example, fluctuations in network traffic may indicate fluctuations in system load. Common indicators for volatility analysis include standard deviation (STD), root mean square (RMS), etc.

[0070] 3.2 Frequency domain feature extraction In addition to time domain features, frequency domain features are also crucial for traffic anomaly identification. Frequency domain features can reveal high-frequency and low-frequency components in the data, helping to detect traffic anomalies caused by rapid changes or emergencies. For example, sudden traffic fluctuations may appear in the high-frequency part of the frequency domain, while periodic or regular fluctuations appear in the low-frequency part. Common frequency domain feature extraction methods include Fourier transform, wavelet transform, etc.

[0071] In this embodiment, frequency domain feature extraction mainly includes the following methods: Fast Fourier Transform (FFT): FFT can convert time series data into frequency domain data, thereby revealing the frequency components in the data. In traffic anomaly identification, the high frequency part of the frequency domain can reflect the instantaneous fluctuations of the network, while the low frequency part can reveal periodic fluctuations. By analyzing the frequency domain characteristics, we can better understand the source of traffic fluctuations.

[0072] Wavelet transform: Wavelet transform is a frequency domain analysis method suitable for non-stationary signals, especially for capturing instantaneous changes in signals. In traffic anomaly detection, the use of wavelet transform can effectively extract high-frequency fluctuation characteristics caused by sudden events such as network attacks and equipment failures.

[0073] 3.3 Multimodal Data Fusion In practical applications, the identification of traffic anomalies usually relies on information from multiple data sources. A single data source often cannot provide enough discriminative information, so multimodal data fusion is needed to combine features from different data sources (such as network traffic data, sensor data, operation logs, etc.) to form a unified feature representation. Multimodal data fusion can not only enhance the ability to identify traffic anomalies, but also effectively improve the robustness of the model.

[0074] In this embodiment, a weighted fusion method is used to perform multimodal data fusion. Specifically, the feature extraction results of each data source are weighted according to their importance to obtain a unified feature vector.

[0075] By weighted fusion of features from different data sources, we can combine the advantages of each data source to obtain a more discriminative feature representation, further improving the accuracy of traffic anomaly detection.

[0076] 3.4 Feature Selection and Dimensionality Reduction In the process of feature extraction, a large number of features may be obtained, many of which may be redundant or irrelevant, which may have a negative impact on the subsequent anomaly detection model. Therefore, it is necessary to select and reduce the dimension of the extracted features to reduce the dimension of the feature space and ensure that the selected features have strong discriminative ability for traffic anomaly identification.

[0077] Commonly used feature selection methods include feature screening based on correlation analysis, information gain, principal component analysis (PCA), etc. By reducing the dimension, the computational overhead can be significantly reduced and the performance of the model can be improved.

[0078] In this embodiment, principal component analysis (PCA) is used to perform dimensionality reduction processing on the extracted features. The goal of PCA is to transform the data into a set of new irrelevant variables (principal components) through linear transformation.

[0079] 4. Traffic anomaly detection Flow anomaly detection is the core part of the present invention, which aims to identify abnormal flow patterns from processed multi-source data. Flow anomalies may be caused by a variety of factors, such as equipment failure, network attacks, or abnormal operating modes. Therefore, the ability to accurately and efficiently detect flow anomalies is of great significance for ensuring the safety, stability, and production efficiency of machine tool operation. In this embodiment, flow anomaly detection uses a machine learning algorithm to identify anomalies by combining time series features, frequency domain features, and multimodal data.

[0080] The main goal of traffic anomaly detection is to distinguish normal traffic from abnormal traffic. By building a reasonable detection model, effective traffic monitoring and alarm can be performed in real-time or batch processing environments. In the present invention, traffic anomaly detection includes model training based on supervised learning, anomaly scoring based on unsupervised learning, and anomaly classification based on hybrid algorithms.

[0081] 4.1 Traffic Anomaly Detection Based on Supervised Learning Traffic anomaly detection methods based on supervised learning rely on historical data sets and labeled normal and abnormal traffic samples for training, and distinguish normal traffic from abnormal traffic by building a classification model. Supervised learning methods can improve detection accuracy and can handle more complex traffic patterns.

[0082] In this embodiment, the anomaly detection method based on supervised learning mainly includes the following steps: Dataset preparation and label construction: First, the feature extraction results from multi-source data are annotated with known normal traffic and abnormal traffic to form a training set. Through historical data analysis, normal and abnormal traffic are manually annotated. After annotation, the dataset is divided into a training set and a test set, where the training set is used for model training and the test set is used to evaluate the performance of the model.

[0083] Feature selection and dimensionality reduction: During the training phase, features that are strongly correlated with traffic anomalies are selected and dimensionality reduction is performed on high-dimensional data to reduce computational complexity and improve the generalization ability of the model. Common feature selection methods include mutual information method and chi-square test method, and dimensionality reduction methods include principal component analysis (PCA) and linear discriminant analysis (LDA).

[0084] Classification model training: Use common supervised learning algorithms, such as support vector machines (SVM), decision trees, and random forests, to build classification models. Support vector machines divide data into normal traffic and abnormal traffic by finding the optimal hyperplane, and decision trees recursively divide the data space to generate a tree structure to achieve classification. Random forests improve classification accuracy through the voting mechanism of multiple decision trees.

[0085] Model evaluation and tuning: The trained model is evaluated by its performance on the test set, using indicators such as accuracy, recall, and F1 value to measure the model's anomaly detection ability. Based on the evaluation results, the model may need to be tuned, including adjusting hyperparameters and selecting more appropriate features.

[0086] 4.2 Traffic Anomaly Detection Based on Unsupervised Learning Traffic anomaly detection methods based on unsupervised learning do not rely on labeled data, but identify abnormal traffic through the inherent laws of the data. Since most traffic data in reality lacks sufficient labeled samples, unsupervised learning methods have good practicality and can adapt to more scenarios.

[0087] In this embodiment, the anomaly detection method based on unsupervised learning includes the following technical means: Cluster analysis: Clustering methods can identify outliers or samples that are significantly different from most data points by clustering similar data points into the same group. Common clustering methods include K-means clustering and DBSCAN clustering. K-means divides data into different clusters by iteratively optimizing the position of the center point, while DBSCAN can automatically discover clusters based on density and identify outliers.

[0088] Density-based anomaly detection: Density-based anomaly detection methods, such as the local outlier factor (LOF) algorithm, calculate the local density of each data point and identify outliers whose density is much lower than the surrounding data points. Points with low density are usually potential outliers.

[0089] Autoencoder: An autoencoder is a neural network model that learns the intrinsic structure of data by compressing the input data into a low-dimensional space and then reconstructing it back to the original input. For normal data, the autoencoder is able to reconstruct the input well, but for abnormal data, the reconstruction error is large. Therefore, by analyzing the reconstruction error, abnormal traffic can be detected.

[0090] 4.3 Anomaly Classification Based on Hybrid Algorithm In order to improve the accuracy of traffic anomaly detection, this embodiment combines supervised learning and unsupervised learning methods to classify traffic anomalies through a hybrid algorithm. The hybrid algorithm takes advantage of the advantages of different methods and can more accurately identify abnormal traffic.

[0091] In the hybrid algorithm, unsupervised learning methods are first used to preprocess and preliminarily classify data into normal traffic and potential abnormal traffic. Then, the potential abnormal traffic is further classified based on supervised learning methods to improve the accuracy of detection.

[0092] process: Use unsupervised methods (such as clustering or density analysis) to perform preliminary detection on the data and extract potential abnormal traffic.

[0093] Use supervised learning methods (such as support vector machines, random forests, etc.) to perform secondary classification on potential abnormal traffic to further distinguish normal traffic from abnormal traffic.

[0094] In this way, the hybrid algorithm can make up for the shortcomings of a single algorithm and improve the robustness and accuracy of traffic anomaly detection.

[0095] 4.4 Anomaly Detection Result Output Finally, the results of traffic anomaly detection will be output to the monitoring system or alarm system. Real-time detection of abnormal traffic can help operators to promptly discover potential network problems or equipment failures, and then take effective measures to repair or adjust them.

[0096] 5. Abnormal alarm and fault location After the flow anomaly detection is completed, the next step is to respond to and handle the abnormal situation in a timely manner through abnormal alarm and fault location. This process is crucial to ensuring the operational safety and production efficiency of CNC machine tools. It can prompt operators of abnormal situations through an effective alarm system and locate the fault, thereby reducing equipment downtime and avoiding more serious damage. The key tasks of abnormal alarm and fault location are to trigger real-time alarms, identify abnormal types, and accurately locate equipment faults based on the results of flow anomaly detection.

[0097] In this embodiment, abnormal alarm and fault location include two main functional modules: alarm mechanism and fault location algorithm. The alarm mechanism sends out an alarm signal and notifies relevant operators based on the abnormal detection results; fault location locates the specific faulty component or fault cause through the analysis of abnormal data, providing data support for maintenance. Through the joint application of these two mechanisms, equipment abnormalities can be accurately responded to in the shortest time.

[0098] 5.1 Abnormal alarm mechanism The goal of the abnormal alarm mechanism is to convert the results of flow anomaly detection into specific alarm signals in a timely manner, ensuring that operators can be informed of the abnormal status of the equipment at the first time. Abnormal alarms can be classified based on different priorities, severity and types to ensure timely response and avoid excessive false alarms.

[0099] In this embodiment, the abnormal alarm mechanism includes the following key steps: Alarm threshold setting: First, set the abnormal threshold based on historical data and the normal working mode of the equipment. When the detected flow index exceeds the preset threshold, the alarm can be triggered. The setting of the threshold takes into account the specific working conditions of the machine tool to avoid frequent false alarms caused by overly strict thresholds. Common threshold setting methods include standard deviation-based settings (for example, an alarm is triggered when the flow fluctuation exceeds 3 times the standard deviation of the mean) or percentage deviation-based settings (for example, an alarm is triggered when the flow exceeds ±10% of the normal value).

[0100] Alarm priority and classification: Different abnormal events may lead to different levels of impact. Setting alarm priority helps to reasonably allocate resources and handle abnormal situations. Alarms are divided into high priority, medium priority and low priority according to the severity of traffic anomalies and possible consequences. For example, when an anomaly that may cause equipment damage (such as a sudden surge in traffic) is detected, a high priority alarm is triggered; when a slight fluctuation is detected, a low priority alarm is triggered.

[0101] Example of alarm prioritization: High priority: The device may fail or shut down.

[0102] Medium priority: equipment performance degradation and inefficiency.

[0103] Low priority: minor fluctuations, no significant impact in the short term.

[0104] Real-time alarm and historical alarm records: After a traffic anomaly is detected, the alarm mechanism will generate an alarm signal in real time and notify the operator through various communication methods (such as email, SMS, system notification, etc.). At the same time, all alarm events will be recorded in the historical alarm database for subsequent analysis and tracking. Historical alarm records help identify long-standing problems and provide a basis for preventive maintenance.

[0105] 5.2 Fault Location The goal of fault location is to analyze the root cause of the anomaly based on the anomaly detection results and determine the specific location of the fault. Accurate fault location can shorten the troubleshooting time, reduce equipment downtime, and improve production efficiency. Fault location usually involves multi-level analysis, including data correlation analysis, fault pattern recognition, and component-level location.

[0106] In this embodiment, fault location is achieved through the following steps: Data correlation analysis: By analyzing the relationship between flow anomalies and other sensor data, a preliminary judgment of the fault mode can be made. For example, a sudden increase in flow may be related to factors such as abnormal motor load and power supply voltage fluctuation. Correlation analysis can effectively narrow the scope of the fault and avoid blindly troubleshooting multiple components of the equipment.

[0107] Data association analysis model: Multivariate regression analysis or correlation analysis can be used to identify the source of anomalies by combining the relationship between flow data and other sensor data (such as temperature, pressure, current, etc.). For example, the relationship between flow and current, if the change trends of the two are consistent, may indicate a motor failure.

[0108] Fault pattern recognition: By analyzing historical fault cases, a fault pattern recognition model is established to help identify possible fault types. For example, if the flow abnormality occurs during a specific operation, it may be a fault in the internal control system of the equipment; and if the abnormal flow occurs when the equipment is started or shut down, it may be an electrical system failure. Common fault modes include mechanical failure, electrical failure, operation error, etc.

[0109] Fault pattern recognition algorithm: Based on machine learning algorithms such as decision trees and support vector machines (SVM), historical fault data can be trained to establish a fault pattern recognition model. When abnormal traffic data is input, the corresponding fault type is identified through the pattern recognition algorithm.

[0110] Component-level fault location: Once the fault mode is determined, the next step is to locate the specific faulty component. By embedding more sensors in the CNC machine tools and obtaining real-time data of key components (such as pumps, motors, hydraulic systems, etc.), the location of the fault can be accurately determined. Through the fault location model, the location of the fault can be further analyzed and maintenance suggestions can be given.

[0111] Fault location model: A Bayesian network-based diagnostic method can be used to combine multidimensional data sources and a known fault mode library to infer the location and specific cause of the fault.

[0112] Visual display and alarm interface: In order to improve the response efficiency of operators, fault location results are usually displayed through a visual interface. Through intuitive graphics and flow charts, operators can quickly view faulty components, abnormality levels and maintenance steps, and optimize the fault response process. Real-time alarm and historical alarm data can also be displayed through charts, dashboards, etc.

[0113] The alarm and fault location system in this embodiment integrates functional modules such as real-time monitoring, anomaly detection, alarm triggering, fault mode recognition and component location. The system can issue an alarm and locate the fault position in the first time based on the detected flow anomaly information, thereby effectively improving the operating reliability and maintenance efficiency of the equipment.

[0114] 6. Online monitoring and model adaptive optimization In the process of identifying flow anomalies in CNC machine tool protocols, real-time and adaptive capabilities are the key to ensuring the long-term stable operation of the system. Online monitoring and model adaptive optimization functions are important components to ensure the accuracy, reliability and adaptability of the detection system. Through online monitoring, the working status and flow data of the machine tool can be tracked in real time, and anomalies can be discovered and processed in time; while model adaptive optimization continuously adjusts and optimizes the flow anomaly detection model through a feedback mechanism to ensure that it can continue to provide efficient detection and prediction capabilities when facing different operating conditions and environmental changes.

[0115] In this embodiment, online monitoring and model adaptive optimization combine real-time data collection, anomaly detection result feedback, and adaptive learning strategies. By monitoring the changes in traffic data in real time, the system can automatically adjust the parameters of the detection model and optimize the algorithm, thereby improving the accuracy and response speed of anomaly detection. The adaptive optimization process is based on real-time analysis of system performance feedback, which enables the model to continuously adapt to new data features and reduce false positives and false negatives.

[0116] 6.1 Online Monitoring Online monitoring refers to the real-time collection and monitoring of the flow data of CNC machine tools during system operation. Through this process, the working status of the machine tool can be grasped in real time, and abnormal flow can be discovered and responded to in a timely manner. Online monitoring can not only monitor the current status, but also record historical data, providing long-term performance analysis and early warning functions.

[0117] In this embodiment, online monitoring mainly includes the following steps: Real-time data collection and transmission: Through the sensor network, the system obtains the flow data of CNC machine tools in real time and transmits it to the central control system. The collection frequency and transmission rate of sensor data need to be optimized according to system requirements to balance data accuracy and processing burden.

[0118] Real-time analysis of traffic data: The online monitoring system analyzes the collected traffic data in real time, using threshold-based monitoring methods or dynamic monitoring methods to detect abnormal fluctuations in the data in a timely manner. Real-time analysis quickly compares the differences between historical data models and current data to discover potential abnormal patterns.

[0119] Real-time alarm and response mechanism: When abnormal traffic data is detected, the system immediately triggers an alarm signal and notifies relevant personnel in a variety of ways. The alarm level is classified according to the severity of the abnormality, and the real-time response mechanism can initiate troubleshooting and recovery operations in a short time.

[0120] Data visualization and interface display: The online monitoring system displays the flow data change trend through a graphical interface, including real-time data graphs, historical data playback, alarm event logs, etc., to help operators intuitively understand the equipment operation status. This interface also supports user-defined views to adapt to different working environments and needs.

[0121] 6.2 Model Adaptive Optimization Since the working state of CNC machine tools is affected by many factors, such as load changes, ambient temperature fluctuations, and different workpiece materials, the original flow anomaly detection model may be unstable or its effect may decrease when facing different operating conditions. In order to improve the adaptability and robustness of the system, the model adaptive optimization adopts an online learning method based on a feedback mechanism to cope with these changes by adjusting model parameters and learning strategies in real time.

[0122] The model adaptive optimization in this embodiment includes the following key steps: Establishment of feedback mechanism: The online monitoring system helps the model judge its performance in actual applications by providing real-time feedback on traffic anomaly detection results. Feedback signals include indicators such as anomaly detection accuracy, false alarm rate, and missed alarm rate. Based on these feedback data, the model can adjust its parameters to improve its adaptability to new data.

[0123] Incremental learning and model updating: To prevent the model from degrading or becoming outdated due to long-term operation, this embodiment adopts an incremental learning strategy. Incremental learning updates the original model parameters by adding new training data in real time to ensure that the model can be optimized according to the new situation. The incremental learning model does not need to be trained from scratch, but is fine-grained based on the existing model to reduce computing overhead.

[0124] Model adjustment strategy: During the model optimization process, the system will automatically select different adjustment strategies according to the changes in data characteristics. For example, when the traffic data fluctuates greatly, the sensitivity of the model can be increased to reduce missed reports; when periodic fluctuations occur, the model can be smoothly adjusted to avoid false positives. Model adjustment strategies include learning rate adjustment, feature selection optimization, and parameter fine-tuning.

[0125] Application of adaptive algorithms: In order to better cope with the dynamic changes of traffic data, this embodiment combines several adaptive algorithms, including adaptive filtering algorithm and adaptive neural network (ANN). Through these algorithms, the system can automatically adjust the detection strategy according to the time-varying characteristics of traffic data, so that the model can maintain efficient detection capabilities in different environments.

[0126] Adaptive Neural Network (ANN) Model: In the initial stage, the neural network will be trained based on historical data to generate initial weights.

[0127] During online monitoring, the model will continuously adjust the network weights based on the newly collected data feedback, thereby achieving an optimized response to real-time data.

[0128] Ultimately, through a series of training and feedback adjustments, the neural network is able to accurately predict traffic anomalies and reduce false positives and missed negatives.

[0129] Implementation of optimization algorithms: In the process of model optimization, strategies such as evolutionary algorithms and particle swarm optimization (PSO) are used to help the model find the optimal parameter configuration by simulating natural selection and global optimization. These optimization algorithms can efficiently search for the most suitable model structure in a complex environment and quickly respond to changes in traffic data.

[0130] 6.3 Performance Evaluation of Adaptive Optimization In order to ensure the effect of adaptive optimization, this embodiment performs regular performance evaluation on the optimized model to analyze the performance of the model under different working conditions. The performance evaluation indicators mainly include: Detection accuracy: The performance of the adaptively optimized model in traffic anomaly identification is evaluated by calculating the accuracy of the model on new data.

[0131] Response time: Evaluate the system's response speed to abnormal events to ensure that online monitoring and alarm mechanisms can be activated in a timely manner.

[0132] False alarm and false negative rates: By comparing the false alarm and false negative rates before and after model optimization, the accuracy improvement of the optimized model can be verified.

[0133] According to the evaluation results, the adaptive optimization strategy is further adjusted to ensure that the model can maintain good performance in various working environments.

[0134] The online monitoring and model adaptive optimization in this embodiment can effectively improve the real-time, accuracy and adaptability of flow anomaly detection. By continuously optimizing the detection model, the system can continue to operate stably under different working conditions, ensuring the efficiency and safety of CNC machine tools.

[0135] Embodiment 1: This embodiment is based on the CNC machine tools of an automobile parts manufacturer. The factory uses multiple CNC machine tools to process various parts, including engine blocks, gears, etc. High-precision control is required during the processing of these parts, and the flow monitoring and control of CNC machine tools is an important factor in ensuring processing accuracy and stable operation of equipment. However, due to the long-term operation of the equipment, the complex working environment and the changing operating conditions, the flow data of the machine tools is prone to abnormalities, which may lead to reduced processing accuracy, equipment failure or even shutdown, thereby affecting production efficiency and product quality. Therefore, it is very important to implement flow anomaly identification and fault location technology.

[0136] In actual applications, these CNC machine tools are connected to a central control system for data collection, real-time monitoring, and flow anomaly identification. The system can monitor key flow parameters such as hydraulic pressure and air pressure inside the machine tool, and conduct real-time analysis and early warning of potential flow anomalies. In addition, the real-time monitoring results of flow anomalies combined with equipment failure information can help technicians quickly locate the source of the failure, thereby optimizing repair and preventive maintenance plans.

[0137] Implementation principle: 1. Multi-source data collection The CNC machine tool in this embodiment is equipped with multiple sensors for monitoring important parameters such as flow, temperature, pressure, load, etc. These sensors collect information such as hydraulic oil flow, gas flow, coolant flow, etc. inside the machine tool in real time and transmit it to the central data processing system. In this system, the flow sensor is responsible for recording the flow of hydraulic pressure, air pressure and coolant during each processing process.

[0138] In this embodiment, the flow sensor of the hydraulic system collects real-time flow data and transmits it to the central control system. At the same time, environmental parameters such as temperature and pressure are also collected synchronously through corresponding sensors. The data of all sensors will be summarized in the data acquisition unit and sent to the data analysis module for processing. Through the collection of multi-source data, the system can more comprehensively grasp the operating status of the equipment and provide reliable data support for subsequent flow anomaly detection.

[0139] 2. Data preprocessing and standardization In order to ensure the accuracy of subsequent data analysis, the system preprocesses and standardizes the collected data. The data preprocessing steps include noise removal, outlier detection and data interpolation to ensure the accuracy and completeness of the data.

[0140] During the data standardization process, various flow data will be uniformly converted according to their units and dimensions. Since flow data may be affected by environmental factors such as temperature and pressure, the system will use a standardized method to convert all flow data into standard units (for example, L / min or m³ / s). In addition, changes in temperature and pressure will also be corrected through temperature compensation and pressure compensation methods to eliminate the impact of environmental changes on flow data, thereby improving data accuracy.

[0141] 3. Data feature extraction and representation In order to perform efficient anomaly identification and fault location, the system needs to extract key feature information from the processed traffic data. In this embodiment, data feature extraction uses time domain analysis, frequency domain analysis and statistical feature extraction methods.

[0142] Time domain feature extraction: By analyzing the time series of traffic data, basic statistical features such as average value, maximum value, minimum value, and fluctuation range are extracted.

[0143] Frequency domain feature extraction: Through Fourier transform and other methods, the flow data is converted to the frequency domain to analyze the frequency components in the flow signal. Abnormal flow often manifests as abnormal fluctuations in a specific frequency band, and these anomalies can be effectively detected through frequency domain features.

[0144] Statistical features: including standard deviation, skewness, kurtosis, etc. of the data, which can help identify potential fluctuation patterns in traffic data.

[0145] The extracted data features will be used in subsequent traffic anomaly detection and fault location models.

[0146] 4. Traffic anomaly detection After feature extraction, the system analyzes the traffic data through the set anomaly detection model to determine whether there is an anomaly. Here, a method based on a combination of threshold and machine learning is used to detect traffic anomalies.

[0147] Threshold-based method: For the standardized flow data, set fixed upper and lower thresholds. For example, the hydraulic oil flow cannot exceed a certain maximum flow value or be lower than a certain minimum flow value. If the flow exceeds or falls below the set threshold, the system will determine it as abnormal.

[0148] Machine learning-based methods: Historical data is trained through supervised learning algorithms such as support vector machines (SVM) or random forests (RF) to build a classification model for normal and abnormal traffic patterns. When new data flows in, the system compares it with the trained model and determines whether the data is abnormal based on the probability value output by the model.

[0149] For example, during a machine tool processing, the hydraulic flow suddenly increases and exceeds the maximum flow value for normal processing, and the system will trigger an abnormal alarm. At this time, the system not only issues an alarm by setting a threshold, but also identifies the potential source of the fault through the learned abnormal pattern.

[0150] 5. Abnormal alarm and fault location When a flow anomaly is detected, the system will trigger an alarm mechanism to notify the operator or maintenance personnel in a timely manner. The alarm mechanism is divided into different levels according to the severity of the anomaly. High-level anomalies will quickly notify relevant personnel and initiate emergency shutdown procedures.

[0151] In terms of fault location, the system combines flow data with other machine tool sensor data (such as temperature, pressure, etc.) for multi-dimensional data correlation analysis. For example, if the hydraulic flow rate increases abnormally and is accompanied by an abnormal increase in temperature, the system may determine that the hydraulic pump is faulty. Through fault pattern recognition, the system can infer specific faulty components, such as hydraulic pumps, pipes, etc., and provide maintenance suggestions.

[0152] 6. Online monitoring and model adaptive optimization To adapt to the complex changes of machine tools under different working conditions, the system will adaptively optimize the flow anomaly detection model. The online monitoring function tracks the machine tool flow data in real time and automatically adjusts the parameters of the detection model according to the equipment status. For example, under certain high-load conditions, the flow fluctuates greatly, and the system will automatically adjust the threshold to avoid false alarms; under low-load conditions, the system will increase the detection sensitivity and reduce missed alarms.

[0153] Model adaptive optimization not only relies on incremental learning of new data, but also includes adjustments based on feedback mechanisms. When the system detects a large number of false positives or false negatives, it adjusts the model parameters by analyzing the false positive and false negative patterns to make the detection process more accurate.

[0154] Through the flow anomaly identification and fault location technology in this embodiment, the CNC machine tool can effectively identify flow anomalies in real-time monitoring and quickly locate the source of the fault. In actual application, the production efficiency of the workshop has been greatly improved, the equipment failure rate has been reduced by 20%, and the machine downtime has been reduced by 30%. In addition, equipment maintenance personnel can obtain fault information in a timely manner through the remote monitoring system, so as to formulate a more scientific preventive maintenance plan, avoid a large number of production interruptions, and ensure the efficiency and stability of production.

[0155] This embodiment demonstrates the technical implementation of abnormal flow identification and fault location for CNC machine tools. Through multi-source data collection, data preprocessing, feature extraction, abnormality detection, alarm and fault location, and model adaptive optimization, the operation safety and production efficiency of CNC machine tools are comprehensively improved.

[0156] This embodiment is based on the CNC machine tools of a mold manufacturing plant, which produces automobile molds and precision molds, etc., involving high-precision processing. CNC machine tools are used to process key components of molds in the production line of the plant, ensuring the high precision and production efficiency of the molds. However, due to the long-term operation of CNC machine tools, the hydraulic system, coolant system, etc. are often affected by wear or other external factors, which can easily lead to abnormal flow, thereby affecting the processing accuracy or causing equipment failure. Therefore, the monitoring and fault diagnosis of abnormal flow of CNC machine tools has become a key link in ensuring the efficient and stable operation of the production line.

[0157] In this scenario, the factory's CNC machine tools need to be hydraulically driven and coolant circulated during each mold processing process. Flow data is continuously monitored and fed back to the central control system in real time. The sensor data of each machine tool is collected and compared with historical working condition data. The system performs anomaly detection based on real-time data analysis. When an abnormal flow is found, the system can not only immediately issue an alarm signal, but also quickly locate the potential source of the fault and reduce production line downtime.

[0158] 1. Multi-source data collection The CNC machine tool in this embodiment is equipped with a variety of devices such as a hydraulic system flow sensor, a coolant flow sensor, a temperature sensor, and a pressure sensor. These devices transmit the collected data to the central control system in real time via wireless or wired means.

[0159] In this embodiment, the hydraulic flow sensor is installed at the oil outlet of the hydraulic pump to monitor the size of the hydraulic oil flow; the coolant flow sensor is installed at the inlet and outlet of the coolant pipeline to monitor the flow change of the coolant; the pressure sensor is installed in the hydraulic system, cooling system and machine tool to monitor pressure fluctuations. Through the joint collection of these multi-source data, the system can fully grasp the working status of the equipment, detect flow anomalies in time and deal with them.

[0160] Specifically, the hydraulic flow sensor and coolant flow sensor monitor the flow of the machine tool hydraulic system and cooling system in real time. When the flow of the hydraulic system is abnormal, such as too low or too high, the system can detect and deal with it in the first time through the collected data.

[0161] 2. Data preprocessing and standardization In order to eliminate noise and interference signals in data collection, this embodiment preprocesses multi-source data. In the data preprocessing stage, high-frequency noise in the sensor signal is first removed by a filtering algorithm (such as median filtering or Kalman filtering). Next, the outlier detection algorithm identifies and removes data that obviously does not conform to physical laws, such as erroneous values ​​caused by sensor failure.

[0162] After data preprocessing, all collected flow data are standardized. The units and dimensions of hydraulic flow data, coolant flow data, and pressure data may be different, so this embodiment uses a unified standardization method to convert all data into the same dimension. For hydraulic flow, the standardized unit is L / min, and for coolant flow, the standardized unit is m³ / s. Through standardization, the system can perform data analysis more uniformly.

[0163] In addition, considering the fluctuation of flow rate under different working conditions, the system also minimizes the impact of environmental changes on flow data through real-time compensation of temperature and pressure. This compensation operation is dynamically adjusted according to real-time data on site to ensure the accuracy and consistency of data.

[0164] 3. Data feature extraction and representation After data preprocessing and standardization, the system enters the feature extraction stage. In this stage, key feature information is extracted from traffic data through time domain and frequency domain analysis methods to facilitate subsequent anomaly detection and fault diagnosis.

[0165] Time domain features: The system calculates the statistical features of the traffic data, such as the mean, maximum, minimum, fluctuation range, and standard deviation. These features reflect the overall trend and fluctuation of the traffic data and can be used to identify basic traffic anomalies.

[0166] Frequency domain features: The flow signal is converted from the time domain to the frequency domain through Fourier transform. Frequency domain features can effectively capture the periodicity or high-frequency disturbances of flow fluctuations and are usually used to identify equipment failures related to flow anomalies.

[0167] Statistical features: Based on historical data, the system also calculates advanced statistical features such as skewness and kurtosis to further capture small fluctuations in anomalies. These features are very important for fine-grained anomaly identification.

[0168] The extracted feature data will be passed as input to the traffic anomaly detection module for subsequent anomaly analysis.

[0169] 4. Traffic anomaly detection The flow anomaly detection in this embodiment adopts a detection model based on machine learning. In this model, the system first uses the traditional threshold method for preliminary screening, and sets upper and lower thresholds for hydraulic flow and coolant flow. For example, if the hydraulic flow is lower than a lower limit or higher than an upper limit, it is determined to be an abnormal state.

[0170] After completing the threshold screening, the system continues to use machine learning-based classification models (such as random forests, support vector machines, etc.) to further determine abnormal situations. By training on historical data, the model can identify which flow fluctuations are within the normal range and which are abnormal fluctuations. For example, the fluctuation range of coolant flow may vary depending on the operating conditions of the machine tool, and the machine learning model can identify and adapt to these changes.

[0171] The model can not only determine whether the flow is abnormal in real time, but also further infer the potential source of the fault based on the characteristics of the abnormal data. For example, if the hydraulic flow suddenly drops and is accompanied by pressure fluctuations, it may be a leak in the hydraulic system. The model will further infer the specific fault location based on these characteristics.

[0172] 5. Abnormal alarm and fault location When a flow anomaly is detected, the system triggers an alarm mechanism to remind operators to take action. The priority of the alarm signal is set according to the severity of the anomaly, and major anomalies are notified to operators through various means such as alarm sound, visual signal or SMS.

[0173] In addition, the system locates faults based on multi-source data. Combining information such as hydraulic flow, coolant flow, pressure data, and temperature data, the system can accurately infer where the fault may have occurred. For example, if the hydraulic flow drops abnormally and the system pressure changes, it may be a hydraulic pump or pipeline failure; if the coolant flow is too low and accompanied by an increase in temperature, it may be a coolant pump failure. Through the fault location module, operators can quickly understand the source of equipment failure, reduce equipment downtime, and perform timely repairs.

[0174] 6. Online monitoring and model adaptive optimization To ensure accurate flow anomaly detection under different working conditions, the system in this embodiment has online monitoring and adaptive optimization functions. The online monitoring function continuously tracks the flow data of the equipment and adjusts the threshold and sensitivity of the flow detection model in real time. For example, during normal operation, the flow fluctuation is small, and the system can automatically lower the threshold to avoid too many false alarms; when the equipment load is large, the fluctuation range of the flow increases, and the system automatically increases the sensitivity to ensure the accuracy of detection.

[0175] In addition, the adaptive optimization of the model can also be adjusted through incremental learning and feedback mechanisms. Whenever the system detects a large number of false positives or false negatives, the model will dynamically adjust according to the feedback data to optimize the detection algorithm, thereby improving the accuracy of traffic anomaly detection.

[0176] Through the flow anomaly identification and fault location technology in this embodiment, the CNC machine tool can monitor key flow parameters such as the hydraulic system and the coolant system in real time during the mold manufacturing process. The system can not only efficiently and accurately identify flow anomalies, but also locate faults by combining multi-source data, issue alarm signals in a timely manner, and provide relevant fault location information, thereby improving the stability and production efficiency of the equipment.

[0177] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying abnormal flow rate for CNC machine tool protocols, characterized in that: The following steps are involved: Step 1: Multi-source data collection: By setting up a data collection module, various types of data are collected from CNC machine tools and their control systems, including network protocol flow data, machine tool sensor data, operation log data, and equipment environment data, and the data format of each data source remains unified; Step 2: Data preprocessing and standardization: Standardize the collected raw data to solve the problems of inconsistent data dimensions, noise interference and missing values, and ensure the alignment of data on the time axis through synchronization algorithms; Step 3: Data feature extraction and representation: Perform feature engineering processing based on the collected data, extract the time series features, frequency domain features and statistical features in the network traffic data, extract the working status features in the sensor data, aggregate all the features to form a high-dimensional feature vector; Step 4: Traffic anomaly detection: The feature vector is input through the deep neural network model, and the spatial features of the data are extracted using a multi-layer convolutional neural network. The long short-term memory network is combined to capture the time-dependent characteristics, and anomaly recognition is performed using an adaptive threshold algorithm.

2. The method for identifying flow anomaly for CNC machine tool protocols according to claim 1 is characterized in that: The convolutional neural network in the deep neural network model is used to extract the local spatial features of the input feature data, and the long short-term memory network is used to extract the time series features of the data, so as to accurately model multi-dimensional and multi-time series data.

3. The method for identifying flow anomaly for CNC machine tool protocols according to claim 1 is characterized in that: The data feature extraction step further includes performing wavelet transform on the network traffic data to convert it into time-frequency features, and performing Fourier transform and sliding average processing on the sensor data to extract frequency domain features and volatility features.

4. The method for identifying flow anomaly for CNC machine tool protocols according to claim 1 is characterized in that: The data preprocessing step also includes denoising the collected data, using wavelet packet transform to remove high-frequency noise, and using Kalman filtering algorithm to smooth abnormal fluctuations in the data.

5. The method for identifying flow anomaly for CNC machine tool protocols according to claim 1 is characterized in that: The flow anomaly detection step further includes using an adaptive threshold method to establish a flow fluctuation model through historical data, and dynamically adjust the anomaly detection threshold by comparing the model output with real-time data.

6. The method for identifying flow anomaly for CNC machine tool protocols according to claim 1 is characterized in that: The flow anomaly detection step also includes using a reinforcement learning algorithm to perform online optimization on the anomaly detection model and dynamically adjust the model parameters to adapt to the flow change characteristics of different machine tools under different loads and operating conditions.

7. The method for identifying flow anomaly for CNC machine tool protocols according to claim 1 is characterized in that: The flow anomaly detection step further includes clustering analysis based on time series data, using a K-means clustering algorithm to classify the flow data, and determining whether an anomaly occurs based on the classification results.

8. The method for identifying flow anomaly for CNC machine tool protocols according to claim 1 is characterized in that: The method also includes real-time monitoring of the machine tool status, triggering an alarm mechanism when a flow anomaly is detected, and outputting detailed fault location information according to different anomaly types.

9. The method for identifying flow anomaly for CNC machine tool protocols according to claim 1, characterized in that: The method can model the fluctuation pattern of network traffic in CNC machine tool protocols, identify abnormal changes in network communication indicators such as data packet size, packet loss rate, delay and bandwidth, and perform correlation analysis with the operating status of the machine tool.

Images