Data acquisition terminal, implementation method, computer equipment and storage medium

Abstract

The invention relates to the technical field of equipment data processing, and discloses a data acquisition terminal, an implementation method, computer equipment and a storage medium, and the data acquisition terminal comprises a data acquisition interface and an edge computing node; the data acquisition interface is used for acquiring multi-source operation state data of acquired equipment from various signal sources; and the edge computing node is used for receiving the multi-source operation state data acquired by the data acquisition interface, performing data analysis, fusion and edge computing on the multi-source operation state data, and performing fault diagnosis prediction and equipment performance analysis on the acquired equipment based on an edge computing result. The multi-source data acquisition requirement is met, the real-time fault diagnosis and prediction requirement of the metallurgical equipment is met, and the problems that the multi-source data acquisition requirement of the metallurgical equipment is difficult to meet, and the real-time fault diagnosis and prediction requirement of the metallurgical equipment is difficult to meet are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of device data processing, and particularly relates to a data acquisition terminal, an implementation method, a computer device and a storage medium. BACKGROUND

[0002] At present, the existing technology for data acquisition research includes:

[0003] 1. Journal literature: Internet of Things Data Acquisition and Processing Based on Edge Computing, which points out that with the development of the Internet of Things, data acquisition and processing face many challenges. The data of the Internet of Things is large in scale and grows rapidly, the data types are diverse and complex, the time and space distribution is uneven, the quality is low and the uncertainty is high, there are also security and privacy problems, real-time requirements and energy consumption problems, etc. For example, it is estimated that by 2025, the number of Internet of Things devices worldwide will reach 75 billion, generating more than 180 ZB of data per day. The Internet of Things data acquisition scheme based on edge computing can effectively cope with these challenges. The edge computing node can be deployed near the data source, its acquisition method needs to be determined according to the data source type and acquisition frequency, etc., and the data privacy and security requirements and scalability need to be considered. At the same time, the data processing method of the edge computing node needs to be determined according to the data type and processing requirements, and also needs to take into account privacy and security, scalability and real-time performance, etc. This literature mainly discusses the Internet of Things data acquisition and processing based on edge computing from a macro perspective, and lacks specific application analysis for specific industrial fields such as metallurgical equipment. For the complex working environment and special data acquisition requirements of metallurgical equipment, the general solution may not be completely applicable. In addition, the specific implementation method and technical details of the edge computing node in the literature are not described in detail, which is difficult to directly guide the actual engineering application.

[0004] 2. Publication number CN110851280A discloses an automated data collection method based on distributed intelligent edge computing technology. The method primarily includes three steps: establishing a terminal management platform, installing terminal devices, and verifying collection results. Specifically, a terminal management platform is established on a physical server to address terminal data transmission and integrated management issues. A wide-angle camera is installed outside the target business system and connected to the terminal device, which is then interconnected with the terminal management platform via a dedicated network. Finally, a list of collected data and a log of data interaction with the remote management platform are displayed on the terminal device. The terminal list, the list of returned data, and related logs are displayed on the terminal management platform. This method provides a non-invasive system connection and data collection method that does not modify existing systems or programs. This method primarily targets data collection for specific business systems and may require extensive adaptation to the diverse sensor types and complex data formats of metallurgical equipment. Furthermore, the connection between the terminal device and the management platform in this method relies on a dedicated network, which can make network deployment difficult and costly in some industrial environments. Furthermore, the data processing and analysis capabilities are relatively weak, making it difficult to meet the real-time fault diagnosis and prediction needs of metallurgical equipment.

[0005] 3. Publication No. CN118689636A discloses an edge computing-based intelligent data acquisition and analysis device and method, primarily used for centralized data collection and storage at oil and gas pipeline construction sites. This device can formulate different priority strategies, early warning analysis strategies, and resource allocation strategies based on different data types, enabling rapid screening, early warning analysis, and coordinated analysis of diverse data. This maximizes the computing efficiency of edge devices and fully leverages the power of central servers. This improves early warning efficiency and accuracy while reducing costs, ensuring safe and orderly construction at oil and gas pipeline construction sites. This application addresses data collection at oil and gas pipeline construction sites, which differs significantly from data collection scenarios for metallurgical equipment. The operating environment of metallurgical equipment is more harsh, placing higher demands on the anti-interference capabilities and reliability of data acquisition terminals. Furthermore, the device and method described in this application fail to fully consider the unique process and operating characteristics of metallurgical equipment, and its data processing and analysis algorithms may not fully meet the needs of metallurgical equipment. Furthermore, the application does not provide a specific solution for integration and collaboration with existing control systems for metallurgical equipment.

[0006] In summary, traditional data acquisition systems often face problems such as data transmission congestion and heavy computing pressure on central servers, resulting in high processing delays and difficulty in meeting real-time requirements. They are not targeted at data collection in specific industrial fields such as metallurgical equipment, and cannot meet the needs of multi-source data collection for metallurgical equipment, making it difficult to meet the needs of metallurgical equipment for real-time fault diagnosis and prediction. Summary of the Invention

[0007] Therefore, the application provides a data acquisition terminal, an implementation method, a computer device and a storage medium to solve the problems that the multi-source data acquisition of metallurgical equipment cannot be met and real-time fault diagnosis and prediction requirements of metallurgical equipment cannot be met.

[0008] In a first aspect, the application provides a data acquisition terminal, which comprises a data acquisition interface and an edge computing node.

[0009] The data acquisition interface is configured to acquire multi-source running state data of an acquired device from multiple signal sources.

[0010] The edge computing node is configured to receive the multi-source running state data acquired by the data acquisition interface, perform data analysis, fusion and edge computing on the multi-source running state data, and perform fault diagnosis and prediction and device performance analysis on the acquired device based on the edge computing result.

[0011] The data acquisition terminal provided by the application has multiple compatible data acquisition interfaces and can be widely applied to various machines and devices to acquire multi-source running state data of the acquired device from multiple signal sources, comprehensively cover electrical, mechanical, thermal and other information of device operation, ensure complete control of device operation status, avoid omission of key information, and have strong universal applicability. The edge computing node can uniformly analyze and fuse data of different signal sources and different formats. After complex and diverse data are converted into a standard format and integrated through a fusion algorithm, the isolated state of the data is broken, and various data are associated and cooperated. This not only enriches the data content, but also significantly improves the data accuracy, lays a solid foundation for subsequent calculation and analysis, and greatly improves the data utilization value. Data calculation is performed at the edge of the device to effectively reduce the time and bandwidth required for data transmission to the cloud or remote server. In a device monitoring scene with strict real-time requirements, data can be quickly processed and analyzed, and the device status can be fed back in a timely manner. With the aid of the edge computing result, the multi-source running state data is deeply analyzed by using a machine learning algorithm or a fault diagnosis model. The abnormal mode in device operation can be detected in advance, and the time and type of potential fault occurrence can be accurately predicted. Key evidence is provided for preventive maintenance of the device to avoid production stagnation caused by device sudden failure, effectively reduce maintenance cost and production risk, ensure smooth and orderly production, meet the multi-source data acquisition requirement of metallurgical equipment and the real-time fault diagnosis and prediction requirement of metallurgical equipment, and solve the problems that the multi-source data acquisition of metallurgical equipment cannot be met and real-time fault diagnosis and prediction requirements of metallurgical equipment cannot be met.

[0012] In an optional embodiment, the data acquisition interface comprises a standard device data interface, a Modbus TCP-based interface, a Modbus RTU-based interface, and an analog and / or switching value interface.

[0013] The standard device data interface is used to connect devices or sensors with Fanuc data interface and / or non-Fanuc data interface;

[0014] Both Modbus TCP-based interfaces and Modbus RTU-based interfaces are used to connect devices or sensors that provide RS485 or RS232 interfaces;

[0015] Analog and / or switching interfaces are used to connect physical quantity monitoring sensors.

[0016] The data acquisition terminal provided by this invention integrates a rich variety of intelligent data acquisition interfaces, including standard device data interfaces, Modbus TCP / RTU interfaces, and sensor interfaces, meeting the data acquisition needs of different types of devices. This eliminates the need for custom development for each device, improving device compatibility and versatility. For sensors without transmitters, transmitters can be flexibly added to convert their output signals into standard electrical signals, further expanding the scope of data acquisition.

[0017] In an optional embodiment, the edge computing node is a Linux-based development board, and the edge computing node includes:

[0018] The data analysis and fusion unit is used to analyze the multi-source operation status data, obtain multiple data features, and fuse the multiple data features based on the data fusion algorithm;

[0019] The real-time analysis unit is used to perform edge computing on the data features after data fusion based on the machine learning algorithm to obtain edge computing results. The edge computing results include fault diagnosis prediction results and equipment performance analysis results.

[0020] This invention provides a data acquisition terminal whose edge computing nodes utilize advanced data analysis algorithms and machine learning techniques to conduct in-depth analysis of device data. This allows for the automatic identification of patterns and trends within the data and the establishment of highly accurate fault diagnosis and prediction models. For example, by applying machine learning to historical data on device operating parameters, the timing, type, and severity of future device failures can be accurately predicted, enabling the implementation of preventive measures, reducing the likelihood of device failures and improving device reliability and production efficiency.

[0021] In an optional embodiment, the edge computing node further includes:

[0022] A local storage unit, used to store multi-source operating status data according to a preset time period;

[0023] The encapsulation and forwarding unit is used to encrypt and authenticate the multi-source operation status data based on the data encryption and encapsulation method, and then encapsulate and forward it.

[0024] The present invention provides a data acquisition terminal, in which local storage not only enables fast access to data and significantly reduces data transmission delays, but also supports continued transmission after power outages, ensuring that data is not lost in unexpected situations such as network outages. When a device generates a large amount of data, key data can be stored in real time for rapid retrieval and analysis, providing strong support for fault diagnosis, predictive maintenance, etc. Locally stored data can also be used to discover trends and patterns in device operation through in-depth mining and analysis of historical data, further improving the level of intelligent device management. This unique combination of local storage and data analysis provides innovative ideas for refined management of equipment. Based on the data encryption and encapsulation method, multi-source operating status data is encrypted and authenticated, then encapsulated and forwarded to prevent data tampering and theft, effectively preventing security issues such as data leakage, tampering, and illegal access.

[0025] In an optional embodiment, the data acquisition terminal also includes an acquisition preprocessing module, which is connected between the data acquisition interface and the edge computing node. The acquisition preprocessing module is used to perform data cleaning, noise removal and data normalization preprocessing on multi-source operating status data.

[0026] The present invention provides a data acquisition terminal, in which an acquisition preprocessing module is connected between a data acquisition interface and an edge computing node. The acquisition preprocessing module is used to perform data cleaning, noise removal and data normalization preprocessing on multi-source operating status data. Through data cleaning, errors, duplications and missing values ​​in the data can be identified and corrected to ensure the accuracy and integrity of the data; noise removal can effectively filter out interference signals, improve the purity of the data, and enable subsequent analysis to better reflect the actual operating status of the equipment; data normalization unifies data of different scales into the same range, eliminates data dimensional differences, enhances data comparability, and provides a strong guarantee for the accuracy of subsequent analysis algorithms. The preprocessed data is more standardized and neat, which can significantly reduce the computational burden of the edge computing node, speed up data parsing and fusion, and optimize the entire data processing flow.

[0027] In an optional embodiment, the data acquisition terminal further includes a communication module and a power supply module;

[0028] The communication module is respectively connected to the data acquisition interface, the acquisition preprocessing module, the edge computing node and the power supply module;

[0029] The communication module is used to provide wired or wireless communication for data transmission between the data acquisition interface and the acquisition preprocessing module, data transmission between the acquisition preprocessing module and the edge computing node, and data transmission between the acquisition preprocessing module and the power module;

[0030] The power module is used to output various voltage levels according to the working status of the data acquisition terminal and automatically adjust the power output;

[0031] The acquisition and preprocessing module is also used to monitor the operating status and parameters of the power module.

[0032] The present invention provides a data acquisition terminal that meets the data transmission requirements in different industrial scenarios through the diversified communication options of the communication module, ensuring the flexibility and reliability of data transmission. By connecting the various key modules of the data acquisition terminal, the communication module builds a complete data transmission link, ensuring the smooth flow of data from raw data acquisition to preprocessing, calculation, and power status monitoring. The multi-level voltage output can accurately adapt to the needs of each module, ensuring that each module can operate under the optimal voltage conditions, improving the working efficiency and stability of the module. The acquisition preprocessing module has the function of monitoring the operating status and parameters of the power module, which not only ensures the safe operation of the power module itself, but also ensures that the entire data acquisition terminal can effectively avoid data loss, equipment damage, etc. when there is a problem with the power supply, thereby improving the reliability and security of the data acquisition terminal system.

[0033] In a second aspect, the present invention provides a method for implementing a data acquisition terminal, which is applied to the data acquisition terminal of the first aspect or any corresponding embodiment thereof, and the method includes:

[0034] Collect multi-source operating status data of the collected equipment from multiple signal sources;

[0035] Receive multi-source operating status data, perform data analysis, fusion and edge computing on the multi-source operating status data, and perform fault diagnosis prediction and equipment performance analysis on the collected equipment based on the edge computing results.

[0036] In an optional embodiment, data parsing, fusion, and edge computing are performed on multi-source operating status data, and fault diagnosis and prediction and equipment performance analysis are performed on the collected equipment based on the edge computing results, including:

[0037] Perform data analysis on multi-source operating status data to obtain multiple data features, and then fuse the multiple data features based on the data fusion algorithm;

[0038] Based on the machine learning algorithm, edge computing is performed on the data features after data fusion to obtain edge computing results. The edge computing results include fault diagnosis prediction results and equipment performance analysis results.

[0039] In a third aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to thereby execute the implementation method of the data acquisition terminal of the above-mentioned first aspect or any corresponding embodiment thereof.

[0040] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the implementation method of the data acquisition terminal according to the first aspect or any corresponding embodiment thereof.

[0041] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions for causing a computer to execute the method for implementing a data acquisition terminal according to the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0043] Figure 1 is a structural diagram of a data acquisition terminal according to an embodiment of the present invention;

[0044] Figure 2 is a structural diagram of another data acquisition terminal according to an embodiment of the present invention;

[0045] Figure 3 This is a schematic diagram of the circuit connection between the data acquisition terminal and the collected device according to an embodiment of the present invention;

[0046] Figure 4 is a workflow diagram for customized data collection requirements according to an embodiment of the present invention;

[0047] Figure 5 is a workflow diagram of data parsing and fusion according to an embodiment of the present invention;

[0048] Figure 6 is a workflow diagram for fault diagnosis of a collected device according to an embodiment of the present invention;

[0049] Figure 7 is a flow chart of the extraction of fault diagnosis features of collected equipment according to an embodiment of the present invention;

[0050] Figure 8is a flow chart of establishing a device fault diagnosis model according to an embodiment of the present invention;

[0051] Figure 9 is a flowchart of a device failure prediction process according to an embodiment of the present invention;

[0052] Figure 10 is a flow chart of establishing a tool fault prediction model according to an embodiment of the present invention;

[0053] Figure 11 is a flowchart of establishing a spindle fault prediction model according to an embodiment of the present invention;

[0054] Figure 12 is a flowchart of establishing a comprehensive fault prediction model according to an embodiment of the present invention;

[0055] Figure 13 is a workflow diagram of device performance evaluation according to an embodiment of the present invention;

[0056] Figure 14 1 is a flowchart of a data encryption workflow based on the MQTT protocol and the SSL protocol according to an embodiment of the present invention;

[0057] Figure 15 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0058] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are 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 those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0059] In the description of the present invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate and simplify the description of the present invention. They are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation, and therefore should not be construed as limitations on the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0060] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "installed," "connected," and "connected" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; internal connections between two components; wireless connections or wired connections. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0061] In addition, the technical features involved in different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0062] In this embodiment, a data acquisition terminal is provided. Figure 1 is a structural diagram of a data acquisition terminal according to an embodiment of the present invention. Figure 1 As shown, the data acquisition terminal includes: a data acquisition interface 201 and an edge computing node 202; the data acquisition interface is used to collect multi-source operating status data of the collected equipment from multiple signal sources; the edge computing node is used to receive the multi-source operating status data collected by the data acquisition interface, perform data analysis and fusion and edge computing on the multi-source operating status data, and perform fault diagnosis prediction and equipment performance analysis on the collected equipment based on the edge computing results.

[0063] Among them, the data acquisition terminal of the embodiment of the present invention is designed with full consideration of the collection of working parameters, operating status, and alarms of various equipment in the factory. Therefore, it is designed with multiple compatible data acquisition interfaces, which can be widely applied to various types of machine tools and equipment. It is used to collect multi-source operating status data of the collected equipment (various types of machine tools and equipment in the metallurgical industry) from multiple signal sources, showing strong universal applicability.

[0064] The edge computing node involved in the embodiment of the present invention uses a high-performance Linux development board, which is responsible for receiving and parsing the pre-processed data, and real-time analysis and processing of edge computing. The edge computing node has super-strong receiving and data fusion functions, and can receive the operating status data of the collected equipment from a variety of signal sources. Whether it is based on the standard equipment data interface, Modbus TCP interface, or RS485, RS232 interface and various sensor interfaces, it can accurately identify and parse the data, fuse data of different formats and sources, and convert it into a unified internal data format, laying the foundation for subsequent processing. The edge computing node can effectively integrate complex and diverse equipment data, break data silos, achieve seamless collaboration between devices, and provide strong data support for comprehensive monitoring and intelligent management of equipment.

[0065] Edge computing nodes integrate advanced data analysis algorithms and machine learning technologies, possessing powerful real-time analysis and processing capabilities. In terms of fault diagnosis, they can establish precise fault diagnosis models. By studying and analyzing massive amounts of historical data, they can accurately identify the data characteristic patterns of equipment under normal operation and fault conditions, promptly detect potential fault hazards, and determine the type and severity of the fault. In terms of fault prediction, they use time series prediction algorithms to accurately predict equipment operating parameters, provide early warnings of possible faults, and provide a strong basis for preventive maintenance. In terms of performance evaluation, device performance indicators can be customized, and the device's performance level can be calculated in real time. By comparing it with historical performance, performance trends can be analyzed to help optimize equipment operating efficiency. This deep integration of data analysis, machine learning, and device management makes edge computing nodes significantly advanced in the field of intelligent equipment management.

[0066] This embodiment provides a data acquisition terminal with multiple compatible data acquisition interfaces, making it widely applicable to various machine tools and equipment. It acquires multi-source operating status data from various signal sources, comprehensively covering electrical, mechanical, thermal, and other aspects of equipment operation. This ensures a complete understanding of equipment operating conditions and avoids missing critical information. The terminal demonstrates strong universal applicability. Edge computing nodes can uniformly analyze and fuse data from different signal sources and formats. After converting complex and diverse data into a standard format, it integrates it through a fusion algorithm, breaking down isolated data and enabling interconnected and synergistic data. This not only enriches data content but also significantly improves data accuracy, laying a solid foundation for subsequent computational analysis and significantly increasing data utilization. Performing data computation at the device edge effectively reduces the time and bandwidth required to transmit data to the cloud or remote servers. In equipment monitoring scenarios with stringent real-time requirements, this terminal can rapidly process and analyze data and provide timely feedback on equipment status. Leveraging edge computing results, machine learning algorithms or fault diagnosis models can be used to deeply analyze multi-source operating status data. This allows for early detection of abnormal patterns in equipment operation and accurate prediction of the time and type of potential faults. It provides a key basis for preventive maintenance of equipment, avoids sudden equipment failures that lead to production stagnation, effectively reduces maintenance costs and production risks, ensures smooth and orderly production activities, realizes the needs of multi-source data collection and meets the needs of metallurgical equipment for real-time fault diagnosis and prediction, and solves the problems of not being able to meet the needs of multi-source data collection of metallurgical equipment and the difficulty in meeting the needs of metallurgical equipment for real-time fault diagnosis and prediction.

[0067] In an optional embodiment, the data acquisition interface includes: a standard device data interface, a ModbusTCP-based interface, a Modbus RTU-based interface, and an analog and / or switch interface; the standard device data interface is used to connect devices or sensors with a Fanuc data interface and / or a non-Fanuc data interface; the Modbus TCP-based interface and the Modbus RTU-based interface are both used to connect devices or sensors providing an RS485 interface or an RS232 interface; the analog and / or switch interface is used to connect physical quantity monitoring sensors.

[0068] like Figure 2 As shown, the data acquisition interface includes:

[0069] (1) Standard equipment data interface:

[0070] Standard equipment data interfaces are Ethernet-based and compatible with a wide range of advanced devices with specific standard data interfaces. Whether you're looking for industry-standard interfaces for common vertical machining centers or specialized standard interfaces for other types of equipment, you'll find the perfect connection solution here. These interfaces adhere to clear data transmission specifications and protocols, ensuring accurate data transfer.

[0071] The data acquisition terminal is equipped with corresponding interface drivers and protocol parsing capabilities for various standard data interfaces. Whether it's a Fanuc or non-Fanuc data interface, it can accurately identify and efficiently process devices. For example, it can accurately collect operating parameters, working status, and alarm information for various machining center equipment, providing a solid data foundation for comprehensive equipment monitoring.

[0072] (2) Network interface based on Modbus TCP:

[0073] It can connect to devices or sensors that provide RS485 or RS232 data acquisition interfaces. The RS485 interface is widely used in many traditional industrial equipment and new sensors. With its strong anti-interference capabilities and long transmission distance, it provides reliable data acquisition in complex industrial environments. The RS232 interface is suitable for short-range data transmission, such as connecting to small devices or instruments, and is flexible and convenient.

[0074] (3) Interface based on Modbus RTU, that is, RS485 or RS232 interface of Modbus RTU:

[0075] Suitable for devices or sensors that provide RS485 or RS232 interfaces. The RS485 interface is more common in some traditional industrial equipment and new sensors. It has strong anti-interference ability and a longer transmission distance, and is suitable for data transmission in complex industrial environments. The RS232 interface is usually used for short-distance data transmission, such as connecting with some small devices or instruments.

[0076] (4) Analog or switch quantity acquisition interface:

[0077] For devices without built-in data acquisition interfaces, the data acquisition terminal of this embodiment provides multiple sensor interfaces, allowing for convenient data collection via the addition of various sensors. Whether it's key parameters like device temperature, pressure, vibration, or other physical quantities that require monitoring, these can all be accurately captured through this data acquisition terminal.

[0078] If the sensor is equipped with an RS485 transmitter, data acquisition is performed using the Modbus RTU interface, demonstrating high compatibility. For sensors without transmitters, data acquisition can be achieved by reading the analog values ​​and switch values ​​of the IOs via the onboard acquisition preprocessing module. Advanced A / D conversion technology is used during analog value reading to ensure data accuracy and reliability. Regardless of the type of device or sensor, the data acquisition terminal of this embodiment can find a suitable interface connection method.

[0079] In an optional embodiment, as Figure 2 As shown, the edge computing node is a Linux-based development board, and the edge computing node includes: a data parsing and fusion unit, which is used to perform data parsing on multi-source operating status data, obtain multiple data features, and fuse multiple data features based on a data fusion algorithm; a real-time analysis unit, which is used to perform edge computing on the data features after data fusion based on a machine learning algorithm to obtain edge computing results, which include fault diagnosis prediction results and equipment performance analysis results. The Linux development board provided in this embodiment adopts dual Cortex-A72 large cores + four Cortex-A53 small cores), a main frequency of 1.8Ghz, a quad-core Mali-T860GPU, 2GB of running memory, and an internal memory eMMC 8G.

[0080] The edge computing node provided by the embodiment has super strong receiving and data fusion functions. Through the data analysis and fusion unit, data from various acquisition and preprocessing modules can be received, whether it is based on a standard device data interface, a Modbus TCP interface, or an RS485, RS232 interface and various sensor interfaces. The data can be accurately identified and analyzed, and data of different formats and sources can be fused and converted into a unified internal data format, laying a foundation for subsequent processing. The edge computing node can effectively integrate complex and diverse device data, break down data silos, achieve seamless collaboration between devices, and provide strong data support for comprehensive monitoring and intelligent management of devices.

[0081] There are many data fusion algorithms, and the embodiment selects different methods according to different devices and data analysis purposes. The data fusion algorithm in the embodiment includes but is not limited to: weighted average method, Kalman filter method, and deep learning data fusion method. The actual application fusion algorithm selects different algorithms according to different actual observation indicators, and each fusion algorithm is further described in detail:

[0082] 1) Weighted average method:

[0083] ① Application scenario: In a vertical machining center, the data of multiple sensors needs to be integrated to evaluate the running state of the machining center, such as judging the tool wear condition through data fusion of vibration sensors, temperature sensors and current transformers.

[0084] ② Data setting: Assuming that the data collected by the vibration sensor, temperature sensor and current transformer are x1=20 (vibration amplitude value), x2=60 (temperature value) and x3=15 (current value), respectively. According to experience, the weights of the vibration sensor, temperature sensor and current transformer are set to w1=0.3, w2=0.3 and w3=0.3, respectively.

[0085] ③ The formula of the weighted average method is:

[0086]

[0087] Substituting the data in ② into the formula, we get: X fusion = 0.3*20 + 0.4*60 + 0.3*15 = 34.5.

[0088] 2) Kalman filter method:

[0089] ① Application scenario: In the motion control of a vertical machining center, the position and speed of the motor need to be accurately estimated, and the data of the encoder and accelerometer are fused.

[0090] ② Data and matrix setting:

[0091] State vector: Let the state vector be where x k is the position, v k is the velocity.

[0092] State transition matrix: Assuming a time interval Δt = 0.1 s, the state transition matrix is

[0093]

[0094] Measurement matrix: The encoder directly measures the position, and the accelerometer indirectly measures the velocity. The measurement matrix is represented as

[0095]

[0096] Process noise covariance matrix:

[0097]

[0098] Measurement noise covariance matrix:

[0099]

[0100] Initial state estimate:

[0101]

[0102] Initial covariance matrix:

[0103]

[0104] Assume that at time k = 1, the encoder measures the position Z11 = 0.5, and the accelerometer measures the velocity Z12 = 2.

[0105] ③ Formula and calculation:

[0106] a. Prediction stage:

[0107] State prediction:

[0108]

[0109] where F0 represents the initial state transition matrix.

[0110] Covariance prediction:

[0111]

[0112] where Q0 is the initial process noise covariance matrix, is the transpose matrix of the initial state transition matrix.

[0113] b. Update stage:

[0114] Calculate the Kalman gain:

[0115]

[0116] Status Update:

[0117]

[0118] Covariance update:

[0119]

[0120] 3) Deep learning data fusion method:

[0121] ① Application scenario: Use data fusion of multiple sensors (such as vibration sensors, temperature sensors, current transformers, piezoelectric sensors, and cameras) to predict the remaining service life of vertical machining center tools.

[0122] ②Data and model settings:

[0123] Data: Collect a large amount of historical sensor data, including vibration signals, temperature values, current values, pressure values ​​and camera image features, and record the corresponding tool life.

[0124] Model: Build a deep learning model using a multilayer perceptron (MLP). Assume the input layer has 5 neurons, corresponding to the data from 5 sensors, the hidden layer has 10 neurons, and the output layer has 1 neuron representing the predicted remaining tool life.

[0125] ③Formula and calculation:

[0126] Forward propagation: Calculation from the input layer to the hidden layer: For the j-th neuron in the hidden layer, its input is:

[0127]

[0128] Among them, w ij is the weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer, x i is the input of the i-th neuron in the input layer (i.e., sensor data), b j is the bias of the jth neuron in the hidden layer. After the activation function is processed, the output of the hidden layer is hj=f(z j ).

[0129] Calculation from hidden layer to output layer: The input of the output layer is:

[0130]

[0131] Where Wj is the weight from the jth neuron in the hidden layer to the output layer, b is the bias of the output layer. The output y of the final output layer is the predicted value of the remaining service life of the tool.

[0132] Backpropagation: Using the mean squared error loss function:

[0133]

[0134] Where N is the number of samples, y n Is the actual remaining tool life, is the predicted value) to measure the difference between the predicted value and the true value.

[0135] Update the model's weights and biases through the backpropagation algorithm, using gradient descent:

[0136]

[0137] Here, α is the learning rate.

[0138] The real-time analysis unit, which is used for edge computing, has a status monitoring and diagnosis process, which is responsible for monitoring, diagnosing and predicting equipment faults; a predictive maintenance process, which is responsible for predicting possible future equipment failures and thus implementing advance maintenance; and equipment performance analysis, which can realize the analysis of equipment production capacity, efficiency, energy consumption, reliability and expected life.

[0139] For example:

[0140] During actual data collection, the data analysis and fusion unit first extracts multiple data features from multi-source operating status data. For example, it extracts features such as vibration frequency, amplitude, phase, temperature change rate, and spindle rotation speed from each collected data. It then combines this data with standard interface data to extract features such as production efficiency and machining accuracy.

[0141] Feature extraction can incorporate the attention mechanism based on deep learning. The specific approach is to first select a suitable base model and then integrate the attention mechanism. Suitable base models include but are not limited to convolutional neural network (CNN) models and long short-term memory (LSTM) models.

[0142] After being processed by the attention network, various data sets are fused according to pre-defined strategies. This fusion strategy can be designed based on signal importance, relevance, or specific domain knowledge. It aims to integrate key information from various signal sources to form a more comprehensive and representative feature vector, which is then used as input for subsequent decision tree models, support vector machines, or random forest regression models to determine the equipment's operating status. For unknown or empirical alarms, a weighted averaging method is used, assigning weights to different signal sources based on their contribution to historical fault data. The weighted signal features are then fused together. For alarms for which the factory already has experience, a rule-based fusion approach is used, combining specific signal features in a targeted manner based on the equipment's operating principles and failure modes. This data fusion effectively lays the foundation for fault diagnosis.

[0143] In an optional embodiment, the edge computing node also includes: a local storage unit for storing multi-source operating status data according to a preset time period; and an encapsulation and forwarding unit for encrypting and authenticating the multi-source operating status data based on a data encryption and encapsulation method and then encapsulating and forwarding it.

[0144] When designing data acquisition terminals, edge computing nodes were equipped with advanced local storage media. High-performance storage media, such as in-memory databases or flash memory, were selected. This not only enables rapid data access and significantly reduces data transmission latency, but also supports continued transmission after power outages, ensuring data is not lost in unexpected situations such as network outages. When devices generate large amounts of data, critical data can be stored in real time for rapid retrieval and analysis, providing strong support for fault diagnosis and predictive maintenance. Locally stored data can also be used to identify trends and patterns in device operation through in-depth mining and analysis of historical data, further enhancing the level of intelligent device management. This unique combination of local storage and data analysis provides innovative ideas for refined device management.

[0145] To realize safe and efficient data interaction with the central server or other devices, the edge computing node is equipped with a packet forwarding unit that adopts a flexible data encryption packaging method. The data encryption packaging method can be encryption authentication through a U disk to ensure the security and authenticity of data transmission, or a TCP+SHA-512 encryption method that uses a powerful hash algorithm to encrypt data to prevent data tampering and theft, or an MQTT (Message Queuing Telemetry Transport) + SSL (Secure Sockets Layer) protocol that ensures accurate packaging of data in the MQTT protocol format and uses the SSL encryption algorithm to encrypt data, effectively preventing data leakage, tampering, and illegal access. The three methods can be selected according to different application scenarios and requirements to meet the diversified data transmission security requirements and reflect the flexibility and adaptability of the edge computing node in data communication security.

[0146] In an optional embodiment, as shown in Figure 2 The data acquisition terminal further includes an acquisition preprocessing module 203 connected between the data acquisition interface and the edge computing node. The acquisition preprocessing module 203 is used for data cleaning, noise removal, and data normalization preprocessing of the multi-source operation state data.

[0147] Specifically, the performance requirements of different devices for the acquisition preprocessing module are fully considered in the design of the data acquisition terminal. Standardized multi-type development boards and data gateway modules, such as small Linux development boards, STM32 small development boards, and serial gateway modules, can be installed on the hardware design. The acquisition preprocessing module can be installed in the device according to the needs, or it can be installed separately on the side of the device to be monitored.

[0148] In an optional embodiment, as shown in Figure 2 The data acquisition terminal further includes a communication module 204 and a power module 205. The communication module 204 is connected with the data acquisition interface 201, the acquisition preprocessing module 203, the edge computing node 202, and the power module 205. The communication module is used for wired or wireless communication between the data acquisition interface and the acquisition preprocessing module, the acquisition preprocessing module and the edge computing node, and the acquisition preprocessing module and the power module. The power module is used to output multiple levels of voltage according to the working state of the data acquisition terminal and automatically adjust the power output. The acquisition preprocessing module is also used to monitor the operating state and parameters of the power module.

[0149] Specifically, the communication module 204 includes a data switch and a wireless communication module. The wireless communication module can be flexibly configured according to the site environment, such as selecting various advanced wireless communication methods such as WIFI, 4G / 5G, and LORA. These are fully considered during the design, and standard modules are uniformly selected to provide installation locations.

[0150] In different application scenarios, these wireless communication methods each have their own advantages. For example, in an industrial field environment with a certain network infrastructure, WIFI can provide high-speed and stable data transmission; for locations where there is no network wiring or wiring is inconvenient, 4G / 5G communication can achieve long-distance, high-mobility data transmission, while LORA excels in low-power consumption and long-distance communication. This diverse wireless communication option can meet the data transmission needs in various complex environments, demonstrating the advanced nature of the communication module. In addition to wireless communication methods, the communication module 204 also supports wired communication. When there is network wiring in the field environment, Ethernet can be selected for data transmission. Ethernet, with its high bandwidth, low latency and stable and reliable characteristics, provides a solid foundation for rapid data transmission. The communication module can automatically switch between wired and wireless communication methods according to actual conditions to ensure the continuity and stability of data transmission, further demonstrating its flexibility and reliability.

[0151] The power module 205 uses a standardized module and is equipped with an emergency battery according to actual needs. A monitoring device is added to enable it to have real-time collection of power data (voltage, current, power consumption), as well as energy-saving control functions. When some equipment is not in use, it can automatically cut off its power supply, thereby achieving the purpose of reducing power consumption and saving energy, while extending the service life of the equipment.

[0152] The power module 205 provides three voltage specifications: 5V, 12V, and 24V. The lithium battery is configured as a 24V10AH battery. The output voltage and current of the power module meet the requirements of the data acquisition terminal and have sufficient stability and reliability.

[0153] The power module 205 can automatically adjust the power output according to the operating status of the data acquisition terminal. For example, when the data acquisition terminal is in sleep mode, the power output is reduced to save energy. When the data acquisition terminal is in operation, the power output is dynamically adjusted according to the load to ensure stable power supply.

[0154] The data acquisition and preprocessing module 203 monitors the status and parameters of the power module 205, such as voltage, current, and temperature. Specifically, the module collects temperature parameters using a temperature sensor and monitors the power module's voltage, current, and other parameters in real time using an AD module. If a power supply anomaly occurs, a warning signal is issued and appropriate protective measures are implemented, such as shutting off the power supply or activating a backup power source, to ensure the safe operation of the data acquisition terminal.

[0155] In practical applications, the data acquisition terminal is connected with various collected device circuits, as shown in the figure, comprising: Figure 3

[0156] (1) Connection with standard device data interface:

[0157] The standard device data interface is divided into Fanuc and non-Fanuc device data interfaces, both of which use different API interface drivers, and the same point is that they both belong to software interfaces and both need to pass through network interfaces for data transmission. The standard device data interface needs to configure the IP addresses of the device side and the acquisition side, requires to be in the same network segment, and can communicate by calling the data interface.

[0158] (2) Connection with interfaces supporting Modbus TCP protocol:

[0159] Some industrial devices support network port Modbus TCP protocol for device data acquisition, require to configure the IP of the device side and the acquisition side, and the IP needs to be in the same network segment. According to the Modbus protocol provided by the manufacturer, the relevant program can be developed for data acquisition.

[0160] (3) Connection with RS485 or RS232 device or sensor interface:

[0161] Traditional industrial devices and most sensors support RS485 Modbus RTU or RS232 protocol, and can be used for device data acquisition. According to the Modbus RTU protocol or RS232 protocol format provided by the manufacturer, the relevant program can be developed for data acquisition.

[0162] (4) Connection with analog or on-off sensor interface:

[0163] Many factory devices do not have or do not open data communication interface, at this time, it is necessary to externally connect mutual inductor, temperature sensor and other sensors. This kind of sensor collects analog quantity, and must pass through IO for A / D conversion to collect data. This kind of sensor can be collected through the AD on the data acquisition pre-processing module 203 board, or can be collected through a dedicated collector. In order to prevent the damage of the board caused by voltage or current during conversion, an isolation optocoupler is used for signal isolation.

[0164] The embodiment also provides an implementation method of the data acquisition terminal, applied to the data acquisition terminal shown in Figure 1 , Figure 2 or Figure 3 , and the method comprises the following steps:

[0165] Step S101, collecting multi-source running state data of the collected device from multiple signal sources.​

[0166] Step S102: Receive multi-source operating status data, perform data analysis, fusion, and edge computing on the multi-source operating status data, and perform fault diagnosis prediction and equipment performance analysis on the collected equipment based on the edge computing results.

[0167] It should be noted that when the data acquisition terminal is working, the edge computing node can configure the parameter file of the data acquisition program to customize user needs. For example, if you want to customize the data collected from the vertical machining center, you can modify the parameter configuration file, select the vertical machining center, and then select the parameters to be collected, the collection frequency and trigger conditions, save the settings, and restart the data acquisition program to achieve the customized collection needs. Step S101 can be a customized data collection requirement workflow, such as Figure 4 As shown, specifically including:

[0168] (1) User Communication: Communicate with users to understand their basic needs and expectations, including the main purpose of data collection, key parameters of the equipment of interest, alarms, etc. Ask users for their specific requirements for data collection, such as which equipment operating parameters, status information, and alarm information they would like to collect. At the same time, understand the user's preferences for sampling frequency, trigger conditions, etc. Then summarize the user's needs and confirm with the user to ensure that both parties have a clear and consistent understanding of the customized data collection requirements.

[0169] (2) Equipment evaluation:

[0170] Conduct an on-site inspection of the user's equipment to understand its hardware characteristics, including interface types and sensor installation locations. Review the equipment's operation logs and maintenance records to understand its historical operating status and common problems. Evaluate the equipment's existing data collection functions and capabilities to determine which data can be acquired directly from the device and which requires external sensors. Analyze the equipment's operating characteristics to identify factors that may affect data collection, such as vibration, temperature, and electromagnetic interference.

[0171] Check the compatibility of the intelligent data collection terminal with the user's device to ensure smooth connection and data collection. Consider any accessories or adapters that may be needed for better compatibility.

[0172] (3) Sensor selection and procurement, testing and rapid development:

[0173] 1) Sensor Selection, Procurement, and Testing: The existing device has five reserved RS485 interfaces, meaning five Modbus RS485 buses. Each bus can connect 16 sensors, totaling 80 sensors. Based on engineering experience, this is sufficient for general equipment monitoring. When existing sensors fail to meet user needs, we conduct an in-depth analysis of the user's desired data type and accuracy requirements. We then determine the required functions and performance indicators for new sensors before selecting, purchasing, and testing them to ensure they can accurately collect the required data.

[0174] 2) Rapid development: If there is no need to select and purchase new sensors, but only the need to develop new acquisition parameters from existing standard data interfaces, or if only a new device is added, the acquisition program can be simply modified and compiled according to customer requirements to adapt to the new acquisition needs.

[0175] (3) Parameter selection:

[0176] 1) Equipment Operating Parameter Selection: Based on user needs and equipment evaluation results, select the equipment operating parameters to be collected. For example, for a vertical machining center, parameters such as spindle speed, cutting force, current, and power consumption can be selected. The importance and priority of each parameter should be determined with the user so that they can be appropriately weighted in data processing and analysis.

[0177] 2) Device status selection:

[0178] Determine the equipment status information that needs to be collected, such as operating status (running, stopped, standby), processing status (processing, idle), maintenance status, etc. Consider the impact of equipment status changes on production and maintenance, and select key status information for collection.

[0179] 3) Alarm information selection: Work with the user to determine the alarm information that requires attention, such as high temperature alarm, excessive vibration alarm, abnormal current alarm, etc. Set the alarm threshold and level so that accurate alarm information can be issued in a timely manner when problems occur.

[0180] 4) Sampling frequency selection: Choose an appropriate sampling frequency based on the importance and rate of change of the parameter. For rapidly changing parameters, such as vibration and current, a higher sampling frequency can be chosen; for slowly changing parameters, such as power consumption, a lower sampling frequency can be chosen. Consider the pressure of data storage and processing, and balance the relationship between sampling frequency and data volume.

[0181] 5) Trigger Condition Selection: Discuss the selection of trigger conditions with the user and determine whether to use time triggering, alarm triggering, or manual triggering based on actual needs. For example, for regularly monitored parameters, you can choose time triggering; for emergency alarms, you can choose alarm triggering; for special testing or debugging scenarios, you can choose manual triggering.

[0182] (4) Modify parameter configuration: Based on the results of device evaluation and parameter selection, determine the configuration files to be modified or added. For example, if a new sensor needs to be added, the configuration file may need to be modified to adapt to the new sensor type. Test and verify the modified accessory files to ensure that they function properly.

[0183] (5) Restart the acquisition program: Before restarting the acquisition program, ensure that all parameter settings and accessory file modifications have been completed and tested. Notify relevant personnel to avoid unnecessary impact on production during the restart of the collected equipment. Restart the acquisition program of the intelligent data acquisition terminal according to the correct steps to ensure that the program can start normally and start collecting data. Check the running status of the acquisition program to confirm whether the data collection is carried out according to the set parameters and trigger conditions.

[0184] (6) Observation and Evaluation: After data collection begins, monitor the collected data in real time. Observe data trends and alarm triggering to ensure the accuracy and reliability of data collection. Address any problems that arise, such as data anomalies and false alarms, promptly.

[0185] Regularly analyze and evaluate collected data to check whether it meets user needs. For example, analyze the equipment's operating status and performance indicators to determine whether parameter settings need to be adjusted or further optimized.

[0186] Based on the results of observation and evaluation, we continuously optimize the parameter settings and accessory files of the intelligent data acquisition terminal, continuously improving the quality and effectiveness of data acquisition to meet the ever-changing needs of users.

[0187] In step S101, through the above-mentioned communication with users, equipment evaluation, sensor selection and procurement, testing and rapid development, parameter selection, modification of parameter configuration, restart of acquisition program, and observation and evaluation, multi-source operation status data of the collected equipment is collected from multiple signal sources.

[0188] Take the vertical machining center as an example to illustrate the workflow of data collection, data analysis and fusion, and edge computing. Figure 5 As shown, the collection of other devices is similar to this and will not be repeated here. Figure 5 It presents the overall workflow from determining the sensor type to setting the acquisition frequency and trigger conditions, and then to data transmission and server data storage. It also briefly describes the data collection, data analysis and fusion, and edge computing of the data acquisition terminal.

[0189] (1) Multi-source data collection, including:

[0190] 1) Determine the sensor type:

[0191] It is assumed that the sensors installed in the vertical machining center include vibration sensors, temperature sensors, current transformers, piezoelectric sensors, and cameras (the types and quantities of actual sensors installed may be different).

[0192] Vibration sensor (accelerometer): installed in key parts of the vertical machining center, such as the spindle and worktable, to monitor vibration during the machining process.

[0193] Temperature sensors: Distributed in spindle bearings, spindle motors, tools, cutting areas, etc., to detect temperature changes.

[0194] Current transformer: installed on the motor power supply line to monitor the current changes of the motor.

[0195] Piezoelectric sensor: can be installed at the contact point between the tool and the workpiece to measure the cutting force.

[0196] Camera: captures image data of the tool appearance.

[0197] 2) Data acquisition frequency setting and trigger conditions:

[0198] Set the acquisition frequency for different sensors based on the VMC's processing characteristics and the need for real-time data. For example, vibration and current sensors can be set to a higher acquisition frequency, such as 100 times per second; temperature sensors can be set to a relatively low acquisition frequency, such as 10 times per second. Select the time trigger condition.

[0199] 3) Multi-source data collection:

[0200] Data such as current, power consumption, power on / off time, CNC model, CNC version number, number of axes currently controlled, spindle number, spindle x-coordinate, spindle y-number, spindle z-coordinate, spindle rotation speed, part number M code, number of parts to be processed, total number of parts to be processed, number of parts required, cumulative power-on time, run time, cutting force, cutting time, and cycle time are collected from the Fanuc standard data interface. Each sensor independently collects vibration, temperature, current, cutting force, and other data at a set frequency and transmits the data to the edge computing node. An example of multi-source data collection based on user-customized requirements is shown in Table 1 below:

[0201] Table 1 Example of multi-source data collection for user customization requirements

[0202]

[0203]

[0204]

[0205] (2) Data preprocessing, including:

[0206] 1) Data cleaning:

[0207] Outlier removal: Using statistical methods or machine learning algorithms, we identify and remove data points that significantly deviate from the normal range. For example, for temperature data, if a measurement exceeds three standard deviations of the normal operating temperature range, it is considered an outlier and removed.

[0208] Noise removal: Use median filtering algorithm to remove noise from sensor data.

[0209] Data normalization: Normalize data from different sensors to have the same dimensions and range. For example, vibration data can be normalized to a value between 0 and 2 (0 represents normal, 1 represents abnormal, and 2 represents an alarm). Temperature data can also be normalized to a value between 0 and 2 (0 represents normal, 1 represents abnormal, and 2 represents an alarm). Current data can also be normalized to a value between 0 and 2 based on the motor's rated current, with 0 representing normal, 1 representing too low, and 2 representing too high. The range of values ​​for the normalized data corresponding to the source data is pre-set and can be continuously adjusted based on model optimization.

[0210] The specific formula used for normalization is:

[0211]

[0212] Among them, X is the original data, X max and X min are the minimum and maximum values ​​of the feature data, X norm is the normalized data.

[0213] (3) Feature extraction and data fusion, including:

[0214] 1) Feature Extraction: Extract features from data from various sensors. For example, from vibration sensor data, features such as vibration frequency, amplitude, and phase are extracted; from temperature sensor data, features such as maximum temperature, average temperature, and temperature change rate are extracted. Feature data is also extracted from multi-source data sets collected through standard interfaces. For example, machining status features can be extracted based on changes in spindle rotation speed, and production efficiency features can be extracted based on data such as the number of parts processed.

[0215] 2) Fusion algorithm selection: There are many data fusion algorithms. This device selects different methods according to different devices and the purpose of data analysis. The fusion methods used in this embodiment are: weighted average method, Kalman filter method, and deep learning data fusion method. The fusion algorithm used in actual application selects different algorithms according to the actual observation indicators. The details of each data fusion algorithm are not repeated here. The comparison of various data fusion algorithms is shown in Table 2 below:

[0216] Table 2 Comparison of various data fusion algorithms

[0217]

[0218]

[0219] (2) Edge computing:

[0220] The data acquisition terminal of this embodiment supports multiple edge computing modes, including a status monitoring and diagnosis process, which is responsible for monitoring, diagnosing, and predicting equipment faults; a predictive maintenance process, which is responsible for predicting possible future equipment failures and thus enabling advance maintenance; and equipment performance analysis, which can analyze the equipment's production capacity, efficiency, energy consumption, reliability, and expected lifespan.

[0221] (4) Data transmission and storage:

[0222] 1) Data transmission: Data processed by edge computing nodes is transmitted to a central server or cloud for further analysis and storage. This can be done using wired or wireless communication methods such as Ethernet, Wi-Fi, and Bluetooth. To ensure the security and reliability of data transmission, one of three encryption technologies and sealing technologies can be used for encryption, encapsulation, and transmission.

[0223] 2) Data Storage: Edge Computing Node Storage (i.e., local storage): Edge computing nodes store data for a short period of time, which can be configured based on user requirements, with a minimum of one week. This stored data is primarily used for real-time monitoring and recent data analysis, enabling rapid response to device status changes and short-term performance evaluation.

[0224] Central server (or cloud) storage: The central server or cloud stores collected data long-term for historical data analysis, trend forecasting, and big data mining. The stored data can be used for equipment performance optimization, production plan adjustments, and decision support.

[0225] Take the vertical machining center as an example to illustrate the fault diagnosis workflow of the collected equipment. Figure 6 The fault diagnosis of other devices is similar to this and will not be described here. Figure 6The equipment fault diagnosis workflow of edge computing nodes is described, and the entire process including data collection, preprocessing, feature extraction, fault diagnosis model establishment, training, evaluation optimization, and fault diagnosis is explained.

[0226] (1) Data collection and preprocessing:

[0227] 1) Multi-source data acquisition through the data acquisition interface: Raw data related to the equipment's operating status is collected from the aforementioned multiple signal sources. This data covers various physical parameters during the equipment's operation. Furthermore, by referencing the equipment's operating time, power-on and power-off times, and combining internal and external data, the equipment's wear and tear curve is simulated in advance. This allows for early warning of equipment risks, proactive maintenance, and prevention of equipment failures.

[0228] 2) Data Preprocessing: The acquisition and processing module preprocesses the collected raw data, including noise removal, filtering, and normalization, to improve signal quality and stability and ensure the accuracy of subsequent processing. For example, a low-pass filter can be used to remove high-frequency noise in vibration signals. Sensor data of varying ranges can be normalized to a uniform range, facilitating subsequent integration and analysis.

[0229] (2) Feature extraction and data fusion:

[0230] 1) Feature extraction: Extract features from the collected data, such as vibration frequency, amplitude, phase, temperature change rate, spindle rotation speed change, etc. Combined with standard interface data, extract production efficiency features, processing accuracy features, etc. Feature extraction can introduce an attention mechanism based on deep learning. The specific method is to select a suitable basic model and then integrate the attention mechanism. The process is as follows: Figure 7 shown.

[0231] A. Appropriate Basic Model: Suppose we collect vibration signal data from a vertical machining center over a period of time and organize it into a three-dimensional tensor of the shape (number of samples, time step, feature dimension). Here, we assume the number of samples is 1000, the time step is 100, and the feature dimension is 1 (considering only one vibration feature), denoted as . Also, the corresponding fault labels are denoted by 0, 1, and 2, which represent three fault categories.

[0232] Convolutional Neural Networks (CNNs): CNNs are suitable for data with local correlations, such as the time-frequency characteristics of vibration signals. The convolutional layers of CNNs can automatically extract local features of the data. For example, when processing the spectrogram of a vibration signal, the convolution kernel can capture correlations between different frequency and time segments.

[0233] Example: Use a convolution layer with a kernel size of 3, a stride of 1, and 16 output channels. Bias For the input sample After the convolution operation, the feature map is obtained The calculation formula is as follows:

[0234]

[0235] Among them, i=0,1,...97 represents the row index of the feature map, and k=0,1,...,15 represents the output channel index.

[0236] Activation layer (ReLU): used to apply the ReLU activation function to the feature map output by the convolutional layer:

[0237]

[0238] Pooling layer (maximum pooling, pooling window size is 2, step size is 2), after pooling, the feature map size becomes 49*16, and the calculation formula is:

[0239]

[0240] Wherein, p=0, 1, ..., 48, q=0, 1, ..., 15.

[0241] Long Short-Term Memory (LSTM) networks: LSTMs can be used to effectively process data with time-series characteristics, such as temperature changes or power output during device operation. They can remember previous state information, enabling a better understanding of the dynamic changes in data.

[0242] The output y after CNN pooling pool As the input of LSTM. Assuming the number of hidden units of LSTM is 32, the input (t represents the time step), hidden state Cell state

[0243] The forget gate formula is:

[0244] f t =σ(W f [h t-1 ,x t ]+b f ) (twenty one);

[0245] in, is the forget gate weight matrix, is the bias, and σ is the Sigmoid function.

[0246] The input gate formula is:

[0247] i t =σ(W i [h t-1 ,x t ]+b i )

[0248]

[0249] in,

[0250] The cell state update formula is:

[0251]

[0252] The output gate formula is:

[0253] o t =σ(W o [h t-1 ,x t ]+b o )

[0254] h t =o t ☉tanh(C t ) (twenty four);

[0255] in, Finally, the hidden state of the last time step is taken as the output of LSTM.

[0256] B. Incorporating attention mechanisms, including:

[0257] Position-based Attention: If the location of the data (such as the time point in a time series) is important for feature extraction, position-based attention can be used. For example, using a vibration signal time series from a device, position-based attention can assign different weights to the likelihood of fault features occurring at each time point. For example, if a fault typically exhibits distinct characteristics in the middle of the vibration signal, position-based attention can assign higher weights to the time points in this middle period.

[0258] The output of LSTM is used as the input of the position attention mechanism.

[0259] The formula for calculating the attention score is:

[0260]

[0261] in, and It is h 49 different elements.

[0262] The formula for calculating attention weight is:

[0263]

[0264] The weighted summation formula is:

[0265]

[0266] Channel-based Attention: When data has multiple channels (for example, in multi-sensor data fusion, where data from different sensors are used as different channels), channel-based attention can determine the importance of each channel for fault feature extraction. For example, when simultaneously monitoring motor vibration and temperature for fault diagnosis, channel-based attention can determine which channel, the vibration channel or the temperature channel, is more critical for feature extraction of a specific fault (such as a bearing fault).

[0267] Assume that the position attention input y pos-att Expand to a feature map of shape (1,32) to apply channel attention.

[0268] The global average pooling formula is:

[0269]

[0270] The formula for calculating the attention weight of the fully connected layer is:

[0271] s = σ(W2ReLU(W1z)) (29);

[0272] in,

[0273] The channel weighting formula is:

[0274] y ch-att =s*y poS-att (30).

[0275] 2) Data fusion: After being processed by the attention mechanism network, various data are fused according to a pre-set strategy. This fusion strategy can be designed based on the importance, relevance, or specific domain knowledge of the signal. It aims to integrate the key information of each signal source to form a more comprehensive and representative feature vector, which is used as input for subsequent decision tree models, support vector machines, or random forest regression models to determine the operating status of the equipment. For unknown or experienced alarms, a weighted average method is used. According to the contribution of different signal sources to the historical fault data, corresponding weights are assigned to them, and then the weighted signal features are fused. For alarms that the factory already has judgment experience, a rule-based fusion method is used. According to the operating principle and failure mode of the equipment, the characteristics of specific signals are targeted and combined. Through data fusion, the foundation for fault diagnosis is effectively laid.

[0276] Assume that there is a simple time-domain statistical eigenvector (such as mean, variance, etc.), and compare it with the channel attention output y ch-att Perform splicing and fusion:

[0277] y fusion =[y ch-att ;Xsta](31).

[0278] (3) Establishment of fault diagnosis model: The fault diagnosis model selects the support vector machine method. The specific implementation process is as follows Figure 8 Shown, including:

[0279] 1) Model Selection and Initialization: Because the relationship between machining center fault type and operating status can be complex, a nonlinear kernel function is preferred. This device uses a radial basis function (RBF) kernel. When initializing the support vector machine (SVM) model, the initial parameters of the kernel function are set. The RBF kernel function requires an initial value, which is a pre-defined range based on past experience diagnosing similar equipment faults or reference to relevant literature.

[0280] The fault type and operating status of the machining center are nonlinear problems. We use the RBF kernel function to map the data into a high-dimensional space. The formula is:

[0281]

[0282] Among them, X i is the i-th sample, X j is the jth sample, σ is the bandwidth parameter of the kernel function, which controls the influence range of the kernel function.

[0283] Substituting the kernel function into the dual problem, we get the dual problem of the support vector machine based on the RBF kernel:

[0284]

[0285] Among them, α i is the Lagrange multiplier, y i ∈{-1, 1} is the corresponding label.

[0286] 2) Model training:

[0287] The preprocessed and partitioned training dataset (the training and test sets can be split 70% to 30%) is fed into an initialized SVM model. During training, the SVM model uses an optimization algorithm to find an optimal decision boundary based on the data's class labels (e.g., normal operation, tool failure, spindle failure, etc.), maximizing the separation of sample points from different fault categories in the feature space.

[0288] Decision function: Solve the dual problem to obtain the optimal Lagrange multiplier After that, the decision function is:

[0289]

[0290] Among them, b* is the bias obtained by calculating the support vector.

[0291] The model constructs a hyperplane that effectively distinguishes various fault types based on the distances between sample points and parameters. The optimization algorithm used is Stochastic Gradient Descent (SGD) and its variant, the Adam optimizer, which adaptively adjusts the learning rate to accelerate model convergence.

[0292] 3) Model Evaluation: Evaluate the trained SVM model using the reserved test set data. Select metrics such as accuracy, recall, and F1-score, which reflect model performance from different perspectives. Input the test set data into the model, obtain the predicted results, and then compare them with the true labels to calculate the values ​​of each metric. For example, accuracy measures the proportion of samples correctly predicted by the model to the total number of samples. A low accuracy indicates that the model may be overfitting or underfitting, and further optimization is needed.

[0293] 4) Model optimization: If the model evaluation results do not meet the requirements, the model needs to be optimized. For SVM model, focus on optimizing kernel function parameters. Use grid search, random search and other methods to fine-tune within the given range. Grid search is to traverse all possible combinations with a certain step size within the parameter range to find the best parameter value for model performance; random search is to randomly select several combinations for trial within the parameter range. At the same time, you can also consider replacing the kernel function, such as trying the polynomial kernel function to see if it can improve the model performance, and keep adjusting until the model reaches the expected diagnostic accuracy and other performance indicators.

[0294] 5) Model application: When the model is trained, evaluated and optimized, and its performance meets the requirements, it will be deployed to the actual fault diagnosis system of the vertical machining center. During the subsequent operation of the machining center, real-time data is collected, features are extracted according to the established process and input into the SVM model, and the model judges whether the machining center has failed and the type of failure based on the input features. Once a failure is predicted, an alarm is sent in time so that maintenance personnel can take appropriate measures to ensure the stable operation of the machining center.

[0295] (4) Real-time fault diagnosis: When the model is available, it can be deployed for real-time fault diagnosis and early warning.

[0296] 1) Real-time data monitoring: Continuously collect the running data of the vertical machining center and transmit it to the edge computing unit for real-time processing. Real-time monitoring of key parameters such as vibration, temperature, current and cutting force, etc. Once abnormal values are found, the fault diagnosis process is triggered immediately.

[0297] 2) Fault diagnosis and early warning: Input real-time data into the fault diagnosis model for rapid diagnosis. If the equipment is found to have a fault, an early warning signal is sent immediately to notify relevant personnel for handling. Provide information such as fault type, location and severity to provide decision support for maintenance personnel.

[0298] (5) Fault repair and optimization:

[0299] 1) Fault repair: Maintenance personnel quickly locate the fault point according to the fault diagnosis result and take appropriate repair measures. After repair, test and verify the equipment to ensure that the fault has been completely eliminated.

[0300] 2) Performance optimization: Analyze the causes of the fault, summarize the lessons learned, and optimize and improve the equipment. According to the fault diagnosis result, adjust the running parameters of the equipment to improve the performance and reliability of the equipment. Update and optimize the fault diagnosis model regularly to improve its accuracy and generalization ability.

[0301] The workflow of equipment fault prediction is illustrated by taking a vertical machining center as an example Figure 9The fault prediction of other equipment is similar to this and will not be described in detail here. Figure 9 The equipment fault prediction workflow of edge computing nodes is described, and the entire process including data collection, preprocessing, feature extraction, fault prediction model establishment, training, evaluation optimization, and fault prediction is explained.

[0302] (1) Data collection and preprocessing:

[0303] 1) Multi-source data acquisition: Data is collected from various sensors of the equipment, covering various physical quantity information such as vibration, temperature, pressure, current, speed, etc. generated when the equipment is running. These sensors are distributed in key parts of the equipment to ensure that the operating status of the equipment can be captured in all directions. The collected data is preliminarily sorted and obviously abnormal or erroneous data points are removed. Let the i-th data point collected by a sensor be Xi. If it satisfies the adjacent data points Xi-1 and Xi+1, (δ is the preset threshold), it is determined to be an outlier and removed. At the same time, the temporal synchronization of different sensor data is ensured for subsequent joint analysis.

[0304] 2) Data preprocessing:

[0305] Data cleaning: remove outliers and noise to ensure data accuracy and reliability. For cutting force data F, if an abnormal peak is generated due to instantaneous electromagnetic shock, the gradient of cutting force change at adjacent sampling points is compared. (Δt is the sampling time interval) and the historical average cutting force of the tool and standard deviation σ F ,when (k is the preset coefficient), it is determined to be a noise point and then removed or corrected using linear interpolation. The linear interpolation formula is: If Fi is missing or an abnormal value, (j is the index of the previous normal data point of i).

[0306] For the spindle vibration data V, when individual discrete ultra-high amplitude vibration values ​​appear, if their duration Δt ab Less than the set 0.1 second and has very low correlation with the vibration data before and after (such as β is a preset threshold), it is determined to be interference noise and processed.

[0307] For missing values ​​caused by interruption of sensor data transmission. For example, if the temperature sensor data is missing at a certain moment, if the temperature changes smoothly before and after the moment, the linear interpolation method can be used to calculate the temperature estimate at the missing moment based on the temperature values ​​Ti-1 and Ti+1 before and after the moment; if the temperature changes are complex during the process switching stage, the temperature average value based on the historical processing stage can be used. filling.

[0308] Data normalization: Normalize different types of data to make them have the same dimension and range. A common normalization method is min-max normalization, as shown in formula (17).

[0309] (2) Feature extraction, including:

[0310] 1) Feature extraction:

[0311] Spindle vibration data feature extraction:

[0312] Time domain characteristics:

[0313] Peak P peak =max(V), effective value Kurtosis Where V is the vibration data sequence, n is the number of data points, is the mean.

[0314] Frequency domain features: Use Fast Fourier Transform (FFT) to convert the vibration signal V(t) into the frequency domain V(f) and extract the amplitude of the specific fault frequency component, such as the characteristic frequency f of the main shaft bearing fault. b and its frequency multiplication kf b (k=1,2,...) there will be a significant increase in amplitude.

[0315] Cutting force data feature extraction: Calculate the ratio of the cutting force components in the X, Y, and Z coordinate axes, such as Rxy = Fx / Fy, Rxz = Fx / Fz, and Ryz = Fy / Fz. This ratio will change when the tool wear is uneven.

[0316] Standard deviation of cutting force fluctuation Under stable cutting conditions, the cutting force fluctuates slightly in the early stage of tool wear, and the standard deviation is at a low value. As the wear intensifies, the cutting force fluctuates more violently, and the standard deviation increases.

[0317] The time-frequency analysis method is used to process the force signal during the tool cutting process to obtain the energy distribution characteristics of the cutting force in different frequency bands. Assume that the force signal F(t) is transformed into a time-frequency distribution S(t,f) after time-frequency transformation. Then the energy of a certain frequency band [f1,f2] is

[0318] Tool image data feature extraction:

[0319] Edge profile features: Accurately depict the tool edge profile and calculate the flatness of the edge through edge detection algorithms such as the Canny operator (L smooth L is the length of the smooth part of the cutting edge. total is the total length of the cutting edge), roughness (h iis the height of a certain point on the cutting edge, is the average height) and other parameters.

[0320] Texture features: Use the gray level co-occurrence matrix P(d,θ) to count the joint distribution of pixel gray levels of the tool surface image in different directions θ and intervals d, and calculate the contrast Correlation Equal texture eigenvalues ​​(μ i 、μ j is the mean, σ i , σ j is the standard deviation).

[0321] Spindle system feature extraction: Form factor CF = Vrms / Vavg (Vavg is the average value of the vibration signal), which reflects the sharpness of the vibration waveform. When the spindle is operating normally, the waveform is relatively smooth and the form factor is close to a certain stable value. When the bearing has problems such as pitting corrosion, the vibration waveform will produce periodic impacts and the form factor will increase.

[0322] Pulse factor IF=P peak / V avg It is more sensitive to monitoring instantaneous impact faults, such as short-term collisions caused by improper assembly of spindle components, and the pulse factor will rise sharply.

[0323] Temperature gradient field characteristics: Multiple temperature sensors are arranged in the axial and radial directions of the spindle bearing. The temperatures of adjacent sensors i and j are T i and T j , then the temperature gradient (d is the sensor distance).

[0324] When the lubrication system is partially blocked, the local temperature of the bearing will rise and the temperature gradient will change abnormally.

[0325] Feed system feature extraction:

[0326] Displacement acceleration characteristics: Based on the feed axis displacement sensor data X(t), acceleration In normal feed motion, the displacement changes smoothly and the acceleration is close to zero. If the screw-nut pair creeps, the displacement acceleration will fluctuate periodically, and the amplitude of the fluctuation is positively correlated with the severity of the creep.

[0327] Fluctuation frequency of the screw axial force: Combined with the pressure sensor data P(t), perform spectrum analysis P(f) (such as FFT) to obtain the frequency spectrum and find the main frequency components of the axial force fluctuation.

[0328] Synchronous error characteristics: When machining complex contour workpieces, multiple axes need to move precisely in coordination. Suppose the actual position of the kth feed axis is The theoretical trajectory is The synchronization error RMS value of synchronization error When the synchronization error exceeds the allowable range of the process, it not only affects the processing accuracy, but may also indicate that there is a fault in the drive motor or control system.

[0329] 2) Feature fusion: Use weighted average method, Kalman filter method, deep learning fusion method and other algorithms to fuse multi-source features to improve the comprehensiveness and accuracy of features. Here, the tool subsystem is calculated using weighted average method. Let the multi-source features be x1, x2, ..., xn, and the corresponding weights be w1, w2, ..., wn The fused features

[0330] The feature layer fusion is currently used. The specific approach is to first extract features from the data of each subsystem separately. The tool subsystem extracts the standard deviation of cutting force fluctuations, time-frequency domain energy characteristics, cutting edge flatness, etc.; the spindle subsystem extracts temperature gradients, vibration waveform factors, etc.; the feed subsystem extracts displacement acceleration, axial force fluctuation frequency, etc. Then, these extracted features are fused according to certain rules to construct a comprehensive feature vector. Dimensionality reduction methods such as principal component analysis (PCA) are used to screen and fuse high-dimensional features of different subsystems. In PCA, let the original feature matrix be X, and its covariance matrix C = 1 / (n-1)*X T X, perform eigenvalue decomposition on C C = UAU T (U is the eigenvector matrix, A is the eigenvalue diagonal matrix), select the eigenvectors corresponding to the first k largest eigenvalues ​​to form the matrix U k , then the feature matrix after dimensionality reduction is Y=XU k .

[0331] (3) Establish a prediction model:

[0332] 1) Model Selection Considerations: For tool failure prediction, given the gradual nature of tool wear and the nonlinear nature of the data, neural network models such as long-short-term memory (LSTM) networks are more suitable. They can learn how multi-source data, such as cutting forces and images, change over time, capturing the dynamic characteristics of the tool from initial wear to failure, and accurately predicting the remaining tool life.

[0333] Support vector machines (SVMs) excel at small-sample, nonlinear classification of spindle system failures, due to their diverse nature, including bearing failure, imbalance, and misalignment. A multi-classification SVM model, leveraging characteristics such as spindle vibration and temperature, categorizes the spindle status into normal, bearing failure, and imbalance failure, enabling rapid and accurate identification of the fault type.

[0334] Considering the comprehensive fault prediction of the entire vertical machining center, ensemble learning methods that combine the advantages of multiple models are highly effective. For example, a combination of random forest and AdaBoost is used. Random forest leverages its parallel computation of multiple decision trees and its strong resistance to overfitting to perform a preliminary classification of various fault characteristics. AdaBoost focuses on misclassified samples in the random forest, iteratively improving the model's accuracy and achieving high-precision predictions for complex fault scenarios in machining centers.

[0335] 2) Model training and optimization:

[0336] When training the LSTM model to predict tool failure, an adaptive learning rate adjustment strategy is adopted, such as learning rate decay. The initial learning rate is set to α0. After m training rounds, the learning rate α m =α0*γ m (γ is the attenuation coefficient, 0<γ<1).

[0337] Introducing regularization terms such as L2 regularization, loss function Where L0 is the original loss function (such as mean square error For life prediction, cross entropy loss for classification tasks), λ is the regularization parameter, W i is the model weight.

[0338] SVM model training and optimization:

[0339] When training the SVM model for spindle fault classification, the cross-validation method is used to select the optimal kernel function parameters and penalty factors. For SVM based on radial basis kernel function (RBF), the kernel function where x i and x j is the sample, and σ is the kernel function parameter. Through grid search combined with fold cross-validation, we traverse different combinations of penalty factors and kernel function parameters to find the parameter configuration that gives the model the highest accuracy on the validation set.

[0340] Integration learning model training and optimization: In the random forest part, each decision tree is trained based on a randomly selected sample subset and feature subset, and the information gain (S is the sample set, A is the feature, The nodes are split based on criteria such as entropy and Pi is the probability of category i.

[0341] During AdaBoost iterative training, the prediction error of the previous model is used (W i,t is the weight of the i-th sample in the t-th round, h t (x i) is the prediction result of the t-th weak classifier, and I is the indicator function) to dynamically adjust the sample weight.

[0342] in Z t is the normalization factor.

[0343] 3) The model establishment process is as follows:

[0344] A. The model building process for tool failure prediction is as follows Figure 10 As shown, Figure 10 This paper describes the workflow for establishing a tool fault prediction model in an edge computing unit, and explains the entire process of establishing, training, evaluating, optimizing, and applying the tool fault prediction model. Specifically, it includes:

[0345] a. Data preparation:

[0346] Collect multi-source data from vertical machining center tools. Divide this data into corresponding datasets according to a certain ratio (e.g., 70% for training, 15% for validation, and 15% for testing), providing foundational support for subsequent model training, tuning, and evaluation.

[0347] b. Model selection:

[0348] Considering that the tool wear process evolves gradually over time and that the cutting force, image and other multi-source data present complex nonlinear relationships, the long short-term memory network (LSTM) was selected to construct the tool failure prediction model. t =σ(W f [h t-1 ,x t ]+b f ), input gate i t =σ(W i [h t-1 ,x t ]+b i ), cell state update Output gate o t =σ(W o [h t-1 ,x t ]+b o ), h t =o t ☉tanh(C t ), h t =o t ☉tanh(C t ), where σ is the Sigmoid function, ⊙ represents element-wise multiplication, W is the weight matrix, and b is the bias term.

[0349] c. Model construction:

[0350] Build an LSTM neural network architecture, which includes an input layer, one or more hidden layers, and an output layer. The input layer receives pre-processed tool cutting force data (such as normalized force value, force component ratio, time-frequency domain features, etc.) and tool image feature data (such as edge profile parameters, texture feature values, etc.). These multi-dimensional data serve as input vectors for neurons. The hidden layer is composed of several LSTM units. The number of hidden layers and the number of units in each hidden layer are determined based on the complexity of the data and computing resources. For example, 2-3 hidden layers are set, with 64-128 units in each layer. Each unit is connected through weights to perform layer-by-layer abstract learning on the input data to mine deep-level tool failure feature patterns. The output layer is set according to the prediction target. For the prediction of the remaining life of the tool, continuous value output is usually used, the number of neurons is 1, and the estimated remaining cutting time or remaining available times of the tool is output; if the tool fault type classification prediction is performed (such as normal, slight wear, moderate wear, severe wear, etc.), the corresponding number of neurons is set according to the number of categories, and the softmax function is used. (Zj is the input of the jth neuron, K is the number of categories) and converts the output into a probability distribution for each category.

[0351] d. Model training: Use the divided training set data to train the LSTM model. Set the initial learning rate, such as 0.001, use the mini-batch gradient descent algorithm, set the number of samples per batch to 32, and input the training data into the model in batches. The model calculates the loss function value between the predicted output and the actual label based on the input tool cutting force, image features, and the corresponding actual tool remaining life or fault type label. For tool remaining life prediction, the mean square error loss function is used. where y i is the actual remaining tool life value, is the model prediction value; for fault classification tasks, the cross entropy loss function is used where y ij is the true label (0 or 1) that the i-th sample belongs to the j-th class, is the probability that the model predicts that the sample belongs to the jth class.

[0352] Then, based on the loss value, the weights between neurons are adjusted layer by layer from the output layer to the input layer through the back propagation algorithm. Taking a simple two-layer neural network as an example (ignoring the bias), assuming that the weights from layer l to layer l+1 are W l , the error signal is δ l+1 , then the weight update formula is △W l =ηδ l+1 (a l ) T , where η is the learning rate, al is the activation value of the first layer. After multiple rounds (e.g., 100-200 rounds) of iterative training, the model gradually learns the inherent mapping rules between tool failure features and remaining life or failure type.

[0353] e. Model optimization: During training, to prevent model overfitting and improve its generalization ability on new data, various optimization strategies are used. On the one hand, L2 regularization is introduced, and the loss function becomes where L0 is the original loss function, λ is the regularization parameter, the weight matrix element of the first layer. When updating the weights, the weight update formula becomes

[0354] On the other hand, adaptive learning rate adjustment strategies are used, such as learning rate decay. Let the initial learning rate be α0, and every m training rounds, the learning rate is updated to where t is the current training round, γ is the decay coefficient (0 < γ < 1). As the training progresses, the learning rate is gradually reduced, allowing the model to make more precise adjustments to the weights in the later training period, avoiding skipping the optimal solution, and ensuring steady improvement in model performance for accurate tool failure prediction.

[0355] f. Model evaluation:

[0356] The validation set data is used to periodically evaluate the model during training. The validation set data is input into the model to obtain the predicted results. For tool remaining life prediction, the mean absolute error (MAE) Root Mean Square Error (RMSE) and other indicators are calculated between the predicted value and the true value to measure the accuracy of the model's prediction; for failure classification tasks, accuracy, recall, F1 value, and other indicators are used to evaluate the model's ability to distinguish different failure types.

[0357] Accuracy

[0358] Recall

[0359] F1 value

[0360] where TP is the true positive, TN is the true negative, FP is the false positive, and FN is the false negative.

[0361] According to the evaluation results, further adjust the model hyperparameters (such as the number of hidden layers, the number of units, the regularization parameter, etc.), optimize the model performance, and after repeated training and evaluation, finally obtain a tool failure prediction model with excellent performance, which is put into practical vertical machining center tool failure prediction applications.

[0362] B. Spindle failure model establishment process Figure 11 As shown, Figure 11 This paper describes the workflow for establishing a spindle fault prediction model for edge computing nodes, and explains the entire process of establishing, training, evaluating, optimizing, and deploying the spindle fault prediction model. Specifically, it includes:

[0363] a. Data collection and processing:

[0364] Collect data from various sensors and standard interface data on key spindle parts of vertical machining centers. Pre-process the collected massive data. Use data cleaning technology to remove abnormal noise points caused by electromagnetic interference, sensor failure, etc. For instantaneous spikes in temperature data T, if ( is the mean temperature, σ T is the temperature standard deviation, k is the preset threshold), it is judged as an outlier and corrected; for discrete outliers in the vibration data V, they are identified and corrected by comparing with adjacent data and combining historical data trends.

[0365] The data normalization method is used to unify data of different dimensions into the same magnitude range. The commonly used normalization formula is: Where x is the raw data, μ is the mean, and σ is the standard deviation. Next, the data is divided into different subsets based on time series or operating condition characteristics, such as normal operation data subsets and various fault data subsets (such as bearing faults, imbalance faults, and misalignment faults). This provides balanced and targeted samples for model training.

[0366] b. Model selection: Given the variety of spindle fault types and the complex boundaries between fault characteristics and normal states, nonlinear support vector machines (SVMs) are a more suitable model choice. SVMs have significant advantages in processing small samples, high-dimensional data, and complex classification problems. They can effectively distinguish different working states of the spindle by constructing the optimal classification hyperplane. In particular, for the linear inseparability that may exist in spindle fault data, the SVM model with radial basis kernel function (RBF) is used. The kernel function The original data can be mapped to a high-dimensional space so that it can be linearly separable in the new space, accurately determining whether the spindle is in a fault state and the specific type of fault.

[0367] c. Model construction: Construct an SVM model architecture based on the RBF kernel function. First, determine the input feature vector of the model and use the pre-processed spindle temperature, vibration amplitude, frequency, and control system related parameters as input. These multi-dimensional features fully reflect the operating status of the spindle. Then, based on experience or cross-validation, preliminarily set the key parameters of the SVM, such as the penalty factor C and the kernel function parameter γ (here The penalty factor is used to balance the complexity of the model and the degree of fit to the training data. If the C value is too large, the model is prone to overfitting, while if it is too small, it may lead to underfitting. γ mainly affects the distribution characteristics after the RBF kernel function maps the data to a high-dimensional space, and plays an important role in the shape of the classification boundary of the model.

[0368] The output of the model is the discrimination result of different spindle fault types, usually in the form of one-hot encoding, corresponding to different categories such as normal, bearing fault, imbalance fault, misalignment fault, etc., and each category corresponds to an output neuron. Calculate the probability that the sample belongs to each type, and then determine the type of spindle fault, where n s is the number of support vectors, α i is the Lagrange multiplier, y i are the labels of the support vectors and b is the bias.

[0369] d. Model training: The SVM model is trained using the divided training set data. Training samples containing various spindle operating condition characteristics are input into the model. The model calculates the distance between the sample and the classification hyperplane based on the input characteristics and the corresponding known spindle fault type label (i.e., the true category of the sample). The optimization goal of the SVM is to solve the dual problem formula:

[0370]

[0371] During the training process, an iterative algorithm is used, such as the Sequential Minimal Optimization (SMO) algorithm, which decomposes the complex quadratic programming problem into multiple easy-to-solve sub-problems, selecting two Lagrange multipliers α each time. i and α j Optimization is performed and model parameters are gradually adjusted so that the classification hyperplane can accurately distinguish different types of spindle faults. After multiple rounds of iterative training, the model learns the inherent mapping rules between spindle fault characteristics and fault types.

[0372] e. Model optimization:

[0373] To improve the performance and generalization ability of the SVM model, a cross-validation combined with grid search method is used to optimize the model parameters. The training set data is further divided into n subsets (usually n is 5 or 10) for fold cross-validation. In each fold validation, one of the subsets is used as the validation set, and the remaining subsets are used as the training set. Grid search is used to traverse different step values ​​(such as from 1 to 10, step 1) or (such as from 0.1 to 1, step 0.1) within a pre-set parameter range, and the validation set accuracy of the model under each parameter combination is calculated. Finally, the parameter combination that achieves the highest validation set accuracy is selected as the optimal parameter of the model, ensuring that the model can accurately identify the fault type when faced with new spindle data.

[0374] f. Model evaluation: Use independent validation set data to conduct a comprehensive evaluation of the trained SVM model. Input the spindle feature data of the validation set samples into the model, obtain the fault type prediction results output by the model, and compare them with the actual fault type of the samples. For classification tasks, the performance of the model is mainly measured by calculating indicators such as accuracy, recall rate, and F1 value (the calculation formula is the same as that for tool failure and is omitted here). Based on the evaluation results, if the model performance does not meet expectations, further adjust the model structure or re-optimize the parameters. After repeated training, optimization and evaluation, a model that can reliably predict spindle failures is finally obtained and applied to the spindle fault monitoring and early warning system of the actual vertical machining center.

[0375] C. The process of establishing a comprehensive fault prediction model for a vertical machining center is as follows Figure 12 As shown, Figure 12 This paper describes the workflow for establishing a comprehensive fault prediction model for edge computing nodes, and explains the entire process of establishing, training, evaluating, optimizing, and deploying the comprehensive fault prediction model. Specifically, it includes:

[0376] a. Data collection and standardization:

[0377] Collect data collected by various sensors and standard interfaces of vertical machining centers. Pre-process the collected massive data. Use data cleaning technology to remove abnormal noise points caused by electromagnetic interference, sensor failure, etc.; use data standardization methods such as Unify data of different dimensions into the same magnitude range for subsequent model processing. Then, divide the data into different subsets based on time series or operating conditions to provide balanced and targeted samples for model training.

[0378] b. Feature extraction: Extract time-frequency domain features from tool data, convert the cutting force signal F(t) into the frequency domain F(f) through fast Fourier transform (FFT), and obtain the energy distribution E(f) = |F(f)| in different frequency bands. 2, high-frequency energy tends to increase when the tool wears; extract image features, use edge detection algorithms (such as Canny operator) to depict the tool edge profile, calculate the edge flatness and roughness, and use the gray-level co-occurrence matrix to count the tool surface texture features.

[0379] In terms of the main shaft, in addition to the conventional vibration time domain characteristics (amplitude, frequency), the waveform factor is introduced Pulse Factor The waveform factor reflects the sharpness of the vibration waveform, and the pulse factor is sensitive to instantaneous impact and is used to monitor bearing pitting, assembly impact and other faults. The temperature gradient field characteristics are constructed from the temperature data. By arranging multiple temperature sensors in the axial and radial directions of the bearing, the temperature difference ΔT=T between adjacent sensors is calculated. i -T j , gain early insight into potential lubrication failures.

[0380] The feed system extracts the displacement acceleration characteristics, assuming the displacement signal is X(t), the acceleration During normal feeding, the displacement acceleration is close to zero, and there will be periodic fluctuations when creeping occurs; analyze the synchronous error characteristics of each axis linkage and calculate the actual position of each feed axis and theoretical trajectory RMS deviation Determine the multi-axis coordination accuracy.

[0381] c. Dimensionality reduction:

[0382] Using principal component analysis (PCA) and other dimensionality reduction techniques, a large number of features extracted from each subsystem are screened. Let the original feature matrix be X, and its covariance matrix Perform eigenvalue decomposition on C: C = UΛU T , where U is the eigenvector matrix and Λ is the eigenvalue diagonal matrix. According to the variance contribution rate of the feature (λ i is the i-th eigenvalue, p is the number of features), select the cumulative contribution rate The first k features of , the feature matrix after dimensionality reduction Y = XU k , where U k It is a matrix composed of the first k eigenvectors, which removes redundant information, reduces the data dimension while retaining key information, and improves the efficiency of subsequent model training.

[0383] d. Model Selection: Given the complexity and diversity of vertical machining center fault types, a single model cannot comprehensively and accurately predict all types of faults. Therefore, an ensemble learning approach was adopted, combining random forest and AdaBoost. Random forest utilizes multiple decision trees in parallel, offering strong processing capabilities for large-scale data and robust overfitting resistance, enabling preliminary classification of fault characteristics. AdaBoost, on the other hand, focuses on samples misclassified by random forest and improves model accuracy through iteration. The complementary strengths of these two approaches make them suitable for complex machining center fault prediction scenarios.

[0384] e-model construction:

[0385] Random forest part: Set relevant parameters. The number of trees N is generally set according to the data size and complexity, such as 100-200 trees, the maximum depth D of each tree is set to 5-10 layers, and the feature ratio p is randomly selected. f Set to 0.6-0.8. Each decision tree is trained based on a randomly selected sample subset and feature subset, and the information gain is calculated based on the information gain. (where S is the sample set, A is the feature, is entropy, p i The nodes are split based on criteria such as (i) and multiple decision trees are grown to output preliminary fault classification results.

[0386] AdaBoost part:

[0387] According to the random forest's misjudgment of the sample, the sample weight is dynamically adjusted. Let the sample weight of the tth round be w t , the error rate of the tth weak classifier Where I is the indicator function, y i is the true label of sample i, h t (x i ) is the prediction result of the tth weak classifier.

[0388] Weights of weak classifiers The sample weight update formula is:

[0389] where Zt is a normalization factor such that Set the sample weight update step size β (e.g., 1.2-1.5), iteratively train M decision trees, gradually improve the overall accuracy of the model, and finally output a comprehensive fault prediction result.

[0390] f. Model training:

[0391] Dataset Partitioning: After collecting and preprocessing, the large amount of data is divided into training, validation, and test sets in a specific ratio, typically 70% training, 15% validation, and 15% test. The training set ensures that it covers data from a variety of normal operating conditions and typical failure modes, providing sufficient material for model learning. The validation set is used to regularly evaluate model performance and adjust model parameters during training. The test set is used to objectively evaluate the model's final performance after training is complete.

[0392] Training Process: Using a pre-defined training set, data covering both normal and various fault samples is input to train a random forest to generate multiple decision trees. Each tree continuously grows by splitting nodes based on the extracted sample and feature subsets, maximizing information gain. AdaBoost calculates sample weights based on the random forest's predictions. The updated samples are then re-input, and subsequent decision trees are iteratively trained, gradually improving the model's ability to identify complex faults. Through multiple rounds of iteration, the model learns the inherent connections between machining center fault characteristics and fault types.

[0393] g. Model optimization:

[0394] Random Forest Optimization: Utilizes out-of-bag (OOB) data to evaluate the performance of random forests. OOB refers to the sample data that each tree does not participate in during training, accounting for approximately one-third of the total sample data. By analyzing metrics such as accuracy and misclassification rate on OOB data, parameters such as the number of trees and the maximum depth of each tree can be adjusted. If increasing the number of trees does not significantly improve OOB accuracy and increases computational cost, the number of trees can be appropriately reduced. If some trees are too shallow, resulting in insufficient classification ability, the tree depth can be appropriately increased to optimize the performance of the random forest model.

[0395] AdaBoost optimization: Closely monitor metrics like accuracy and recall on the validation set, and fine-tune parameters like the sample weight update step size and the number of weak classifiers. If validation set accuracy improves slowly with increasing iterations, increase the sample weight update step size appropriately to focus subsequent decision trees on difficult-to-classify samples. If an excessive number of weak classifiers leads to overfitting, reduce the number appropriately to ensure that the AdaBoost model maintains good generalization while improving accuracy.

[0396] h. Model evaluation and application:

[0397] Model Evaluation: Using an independent validation set, we input multi-source feature data from the validation set samples into the model, obtain the model's fault prediction output, and compare it with the actual fault types in the samples. We calculate metrics such as accuracy, recall, and F1 value. Accuracy reflects the proportion of samples correctly predicted by the model out of the total number of samples. Recall measures the model's ability to correctly predict a specific fault type. The F1 value comprehensively considers both accuracy and recall to provide a more balanced performance evaluation. Based on the evaluation results, we determine whether the model's performance meets expectations.

[0398] Model Application: If the model meets performance requirements, it is deployed on-site at a vertical machining center (VMC). Connected to the machining center's control system via distributed edge computing nodes, the model collects data and runs the model in real time, providing real-time monitoring and early warning of equipment failures. If the model predicts an impending failure, such as a tool's remaining life falling below a set threshold, abnormal spindle vibration, or a failure probability exceeding a safe range, an alert is immediately issued to the operator or production management system, prompting appropriate measures such as premature tool replacement or spindle maintenance to ensure stable operation and high production efficiency.

[0399] (4) Real-time fault prediction:

[0400] 1) Real-time data monitoring: Continuously collects operating data from the vertical machining center and transmits it to the edge computing unit for real-time processing. Key parameters such as vibration, temperature, current, and cutting force are monitored in real time. If any abnormal trends are detected, the fault prediction process is immediately triggered.

[0401] 2) Fault Prediction and Early Warning: Real-time data is fed into a fault prediction model to generate real-time predictions. If a failure is predicted within the next period of time, an early warning signal is immediately issued to notify relevant personnel for action. Information such as the type of fault, the likely time of occurrence, and severity is provided to inform maintenance decisions.

[0402] (5) Maintenance decision-making and optimization:

[0403] 1) Maintenance Decision-Making: Develop a reasonable maintenance plan based on the fault prediction results. For example, if severe tool wear is predicted, schedule tool replacement in advance; if a high risk of spindle failure is predicted, perform maintenance in advance. Determine the optimal maintenance timing and method based on the equipment's production plan and maintenance costs.

[0404] 2) Performance Optimization: Analyze fault prediction results, identify potential issues in equipment operation, and optimize performance. Adjust equipment operating parameters, such as cutting speed and feed rate, to improve production efficiency and processing quality. Regularly update and optimize the fault prediction model to improve its accuracy and adaptability.

[0405] Equipment performance evaluation workflow:

[0406] Taking the vertical machining center as an example, the workflow of the data acquisition terminal to evaluate the performance of the collected equipment is explained. Figure 13 As shown, Figure 13 This article describes the performance evaluation workflow for edge computing nodes, covering the entire process from data collection and preprocessing to metric calculation, performance evaluation, performance optimization, and continuous monitoring. The performance evaluation of other devices is similar and will not be discussed further here.

[0407] (1) Data collection:

[0408] 1) Multi-source data acquisition: Vibration sensors, temperature sensors, current sensors, and force sensors installed on the vertical machining center collect physical parameters during operation. Data such as current, power consumption, power on / off time, CNC model, CNC version, number of axes currently controlled, spindle number, spindle x-coordinate, spindle y-coordinate, spindle z-coordinate, spindle speed, part count M-code, number of parts being machined, total number of parts being machined, number of parts required, cumulative power-on time, run time, cutting force, cutting time, and cycle time are collected through the Fanuc standard data interface.

[0409] 2) Data preprocessing:

[0410] Data cleaning: Remove outliers and noise to ensure data accuracy and reliability. For example, for temperature data, if a measurement value exceeds three standard deviations of the normal operating temperature range, it is considered an outlier and removed.

[0411] Data normalization: Normalize different types of data to make them have the same dimension and range to facilitate subsequent analysis.

[0412] (2) Performance indicator determination stage:

[0413] 1) Machining accuracy evaluation indicators:

[0414] Deviation between actual processing size and design size: Calculate the deviation value by measuring the size of the processed part and comparing it with the design size.

[0415] Dimensional stability: Process the same part multiple times, analyze the dimensional fluctuations, and evaluate the stability of the machining accuracy.

[0416] Surface roughness: Use a surface roughness measuring instrument to measure the roughness of the machined surface and evaluate the machining quality.

[0417] 2) Production efficiency evaluation indicators:

[0418] Number of parts processed per unit time: Calculate the output per unit time based on the total number of parts processed and the running time.

[0419] Cutting efficiency: The product of cutting force and cutting speed, which evaluates the efficiency of the cutting process.

[0420] Tool change time: Count the number of tool changes and tool change time during the machining process to evaluate the impact of tool changes on production efficiency.

[0421] 3) Energy consumption evaluation indicators:

[0422] Power consumption: Get the power consumption of the device directly from the collected data.

[0423] Energy utilization rate: The ratio of the number of processed parts to power consumption is used to evaluate the energy utilization efficiency of the equipment.

[0424] 4) Reliability evaluation indicators:

[0425] Mean Time Between Failures (MTBF): Counts the normal operation time and failure times of the equipment to calculate the mean time between failures.

[0426] Mean Time Between Failures (MTTR): Statistics on the repair time after equipment failure and calculate the time between failures.

[0427] Equipment availability: The ratio of equipment normal operation time to total operation time, which evaluates the reliability of the equipment.

[0428] (3) Performance index calculation:

[0429] 1) Machining Accuracy Assessment: Calculate the mean and standard deviation of the deviation between the actual machining dimensions and the designed dimensions to assess the accuracy and stability of machining accuracy. Analyze the dimensional deviation changes under different machining conditions to identify factors affecting machining accuracy.

[0430] 2) Production Efficiency Assessment: Calculate the number of parts processed per unit time and cutting efficiency to evaluate the production efficiency of the equipment. Analyze the impact of tool change time on production efficiency and provide recommendations for optimizing tool change strategies.

[0431] 3) Energy Consumption Assessment: Calculate the equipment's power consumption and energy utilization, assess the equipment's energy consumption, analyze energy consumption changes under different processing tasks, and provide recommendations for energy-saving optimization.

[0432] 4) Reliability Assessment: Calculate mean time between failures, time between failures, and equipment availability to assess equipment reliability. Analyze the causes and frequency of failures and provide recommendations for improving equipment reliability.

[0433] (4) Performance analysis and reporting:

[0434] 1) Performance Analysis: Analyze calculated performance indicators to understand the performance of the device. This can be done by plotting charts, performing trend analysis, and comparing performance over different time periods or across different devices to identify performance trends and potential issues.

[0435] For example, you can plot a curve of equipment performance indicators changing over time to observe the stability and trend of performance; you can compare the performance indicators of different devices to find out which devices have better or worse performance; you can analyze the relationship between performance indicators and other factors (such as equipment operating parameters, environmental conditions, etc.) to find out the key factors affecting performance.

[0436] 2) Generate performance report:

[0437] 1) Data visualization: The collected data and performance evaluation results are displayed in the form of charts, such as bar charts, line charts, scatter plots, etc., to facilitate intuitive understanding of the performance status of the equipment.

[0438] 2) Evaluation Report Generation: Based on the performance evaluation calculation results, a detailed performance evaluation report is generated, including evaluation results in terms of machining accuracy, production efficiency, energy consumption, and reliability. The report should provide improvement suggestions and optimization measures to provide a reference for equipment management and maintenance.

[0439] (5) Continuous monitoring and optimization:

[0440] 1) Continuous Monitoring: Device performance evaluation is an ongoing process that requires continuous monitoring and evaluation of the device. Edge computing devices can regularly collect data, update performance indicator calculations, and perform performance analysis.

[0441] Through continuous monitoring, changes and problems in equipment performance can be discovered in a timely manner, and corresponding measures can be taken to adjust and optimize.

[0442] 2) Optimization measures: Based on the results and analysis of the performance evaluation, take appropriate optimization measures to improve the performance of the equipment. This may include adjusting the equipment's operating parameters, performing equipment maintenance and servicing, and upgrading the equipment's hardware or software.

[0443] The implementation of optimization measures should be based on the actual situation of the equipment and the needs of users, and an effect evaluation is required to ensure the effectiveness of the optimization measures.

[0444] Data encryption and packaging process of the data collection terminal during data transmission:

[0445] Choose MQTT+SSL protocol. The combination of MQTT and SSL protocol is not a simple superposition, but a unique technical means to achieve stronger data security protection, such as Figure 14 As shown, Figure 14This paper describes the data security workflow of edge computing nodes, and explains the entire process of certificate generation, certificate and key configuration, connection verification, encrypted transmission, and integrity verification. The implementation process is as follows:

[0446] 1) Generate a digital certificate

[0447] Server certificate generation:

[0448] This algorithm utilizes an innovative key generation algorithm based on elliptic curve cryptography (ECC) and improved quantum-resistant algorithms. By incorporating dynamic random parameters into the selection of elliptic curves, the generated public and private key pairs remain highly secure even in the face of quantum computing attacks. Furthermore, a multi-factor hash function is used to process the random number seed, increasing the complexity of the key.

[0449] The core steps of the algorithm are:

[0450] (1) Elliptic curve parameter selection:

[0451] Curve selection: First, select a suitable elliptic curve E from the standard elliptic curve set (such as the curve recommended by NIST). The curve is defined on the finite field GF(p) and its equation form is y 2 =x 3 +ax+b(mod p), where a,b∈GF(p) and To increase the randomness and security of the curve, we can introduce additional randomization parameters to fine-tune the curve equation.

[0452] Base point selection: Choose a base point G on the elliptic curve with order n, i.e. nG = O (O is the point at infinity of the elliptic curve). The base point must be selected to ensure that its order n is a large prime number to ensure the difficulty of the discrete logarithm problem.

[0453] (2) Introduction of quantum-resistant algorithms:

[0454] Quantum random number generation: A quantum random number generator (QRNG) is used to generate true random numbers. Based on the uncertainty principle of quantum mechanics, quantum random numbers are unpredictable and truly random. A QRNG is used to generate a random number r1, which is used in the subsequent key generation process.

[0455] Lattice Reduction Algorithm Integration: Lattice reduction algorithms from lattice-based cryptography are introduced to process the random number r1 to generate a new random number r2. Specifically, r1 is represented as a vector in the lattice. The lattice reduction algorithm transforms this vector to obtain a more secure vector, the corresponding integer of which is r2.

[0456] (3) Private key generation:

[0457] Random number combination: The quantum random number r2 is combined with the random number selected during traditional ECC private key generation. The traditional ECC private key d is a randomly selected integer in the interval [1, n-1]. We use a specific combination function f(r2) to map r2 to the interval [1, n-1], generating a new random number d'. The combination function f is a SHA-256 hash function. It takes r2 as input, performs a hash operation modulo n, and obtains d' = SHA256(r2) mod n.

[0458] Private key determination: The final private key d is determined by traditional random selection and d' through weighted summation, that is, d = α*d random +(1-α)*d', where d random It is an integer randomly selected during the traditional ECC private key generation process. α is a weight coefficient with a value range of [0,1] and can be adjusted according to actual security requirements.

[0459] (4) Public key generation:

[0460] Elliptic Curve Multiplication: According to the principles of ECC, the public key Q is obtained by performing elliptic curve multiplication on the private key d and the base point G, that is, Q = dG. During the calculation process, an efficient elliptic curve multiplication algorithm is used to improve computational efficiency.

[0461] To generate digital certificates, we utilize zero-knowledge proof technology combined with an innovative multi-party secure computation approach. First, the server encrypts and combines its detailed information (including server identity, network address, and service scope) with dynamic environmental parameters (such as current server load and real-time network topology), along with a millisecond-accurate timestamp. Using zero-knowledge proof technology, the authenticity and integrity of this information are proven to the certificate generation module without leaking sensitive information. Then, utilizing the principles of multi-party secure computation, this encrypted information is distributed to multiple distributed nodes for collaborative computation, generating a unique digital certificate.

[0462] When generating a Certificate Signing Request (CSR), in addition to conventional information, the aforementioned dynamic environment parameters and timestamp are also incorporated. The CSR is then sent to a rigorously screened and optimally configured Certificate Authority (CA). This CA employs a multi-factor verification mechanism that not only rigorously verifies the authenticity of the information provided by the server but also uses blockchain technology to trace the server's historical security records, ensuring the high reliability and authority of the issued digital certificate.

[0463] Optionally, if client certificate generation is required, the following method is used: If two-way authentication is required, the client generates a digital certificate using an innovative approach based on a combination of homomorphic encryption and attribute encryption. Using a specially developed certificate generation tool, a public-private key pair and a CSR are generated based on specific attributes such as the client's hardware fingerprint, software environment information, and user-defined security policies.

[0464] For special scenarios, a unique self-signed certificate generation system has been developed. This system is built on blockchain distributed ledger technology and the threshold signature algorithm. When a server, acting as a CA, issues a certificate to a client, it first leverages the blockchain's immutable nature to record the entire certificate generation process, including the client's application information and the server's review process. Furthermore, using the threshold signature algorithm, the signature private key is split into multiple shares, shared by multiple server nodes. Only when a certain number of nodes sign together can the certificate be signed, ensuring its authenticity and integrity.

[0465] 2) Configure the MQTT server for the data acquisition terminal:

[0466] Installing an SSL / TLS library: The MQTT server installs an optimized and customized SSL / TLS library, tuned for the server's specific architecture and performance requirements. The Mosquitto server configuration file is deeply customized. When enabling SSL / TLS, not only does it specify the certificate and key paths, but it also sets a series of advanced security parameters. An adaptive dynamic key update algorithm dynamically adjusts the key update frequency based on server load and network security threats. Adaptive encryption strength adjustment technology automatically adjusts encryption strength based on the security requirements of connecting clients and the server's own computing power.

[0467] Configure certificates and keys: Configure the server's specially encrypted digital certificate and private key to the MQTT server. This is done using dual encryption technology based on fully homomorphic encryption and secret sharing. During configuration, the encrypted certificate and private key are stored in ciphertext in a secure storage area on the server. A secret sharing algorithm is used to split the decryption key into multiple shares, each stored on different storage media.

[0468] Fully homomorphic encryption, including:

[0469] Key generation: Generates a public key pk and a private key sk. The public key is used to encrypt data, and the private key is used to decrypt data. Encryption: Use the public key pk to encrypt the plaintext m, resulting in the ciphertext c = Enc(pk,m).

[0470] Homomorphic computation: operations such as addition and multiplication can be performed on the ciphertext c: for example, c1+c2=Enc(pk,m1+m2), c1*c2=Enc(pk,m1*m2), where c1=Enc(pk,m1) and c2=Enc(pk,m2).

[0471] Decryption: Use the private key sk to decrypt the calculated ciphertext and obtain the plaintext result Dec(sk,c'), where c' is the ciphertext after homomorphic calculation.

[0472] Secret sharing, including:

[0473] Secret Splitting: Split the original secret S into n shares s1, s2, ... sn. A threshold t (1 ≤ t ≤ n) is usually specified, indicating that at least t shares are required to recover the original secret.

[0474] Share distribution: Distribute t shares to n different participants.

[0475] Secret recovery: When at least t participants provide their shares, the original secret can be recovered using the corresponding recovery algorithm.

[0476] When the server starts, it uses an innovative key loading mechanism:

[0477] When the server boots up, it utilizes a key loading mechanism based on a Trusted Execution Environment (TEE) and federated learning. First, the TEE ensures the security of the key loading process, preventing malicious theft or tampering. Then, using federated learning technology, it obtains shares of the decryption key from multiple secure nodes and performs collaborative decryption within the TEE, ensuring rapid key availability and security. Furthermore, real-time monitoring and automatic update technologies ensure the validity of certificates and keys.

[0478] Set security policies: Develop highly personalized security policies, combining machine learning algorithms and risk assessment models to select the most appropriate encryption algorithm for the current environment. A dynamic decision-making mechanism is used to determine whether to require client authentication, adjusting it in real time based on factors such as the client's trust level, connection history, and the current network environment. Allowed cipher suites can be specified, and client certificate verification methods can be configured, such as using blockchain-based distributed ledger technology to verify certificate validity and issuing authority. Smart contracts can be used to automate the verification process, improving efficiency and accuracy.

[0479] 3) Configure the MQTT client:

[0480] Importing CA certificates: The client uses an innovative certificate import method based on blockchain and differential privacy. This not only imports trusted CA certificates but also verifies and updates them in real time. If using a certificate issued by a public CA, the client utilizes an intelligent certificate detection system, combined with blockchain's distributed ledger technology, to automatically identify and update the built-in root certificates of common CAs. Differential privacy technology protects the client's privacy during the verification and update process, ensuring the validity of the certificate. For self-signed certificates, the server's CA certificate is manually imported into the client's trust store via a secure transmission channel, and cryptographic verification technology is used to ensure the certificate's integrity and authenticity.

[0481] Configure client certificates and keys: If client authentication is required, the client configures a specially encrypted and protected digital certificate and private key. This is protected by dual encryption techniques based on attribute encryption and homomorphic encryption. When connecting to the MQTT server, the client uses a secure certificate transmission protocol to send its certificate to the server for authentication, preventing the certificate from being stolen or tampered with during transmission.

[0482] Setting connection parameters: In the client code, when setting the MQTT server address, port number, and parameters related to enabling the SSL / TLS connection, dynamic configuration technology is used to adjust connection parameters in real time based on the network environment and server status. The SSL / TLS version and encryption algorithm used are specified to ensure that the client and server configurations match. An intelligent negotiation mechanism is used to automatically adjust parameters during the connection process to adapt to different network conditions and security requirements.

[0483] 4) Establish connection and handshake

[0484] Client-Initiated Connection: When a client sends a connection request to an MQTT server, it uses an encrypted request packet containing information such as the SSL / TLS version and encryption algorithm supported by the client, along with the client's authentication information and security token to enhance the security of the connection request. Zero-knowledge proof-based authentication technology is used to prove the client's identity to the server without leaking sensitive information.

[0485] Server Response: After receiving the connection request, the server selects the most appropriate encryption algorithm and SSL / TLS version based on the information provided by the client and sends the server's digital certificate to the client. The certificate transmission process uses encrypted transmission and digital signature technology to ensure the authenticity and integrity of the certificate.

[0486] Client Verification of Server Certificate: After receiving the server certificate, the client uses a multi-factor authentication mechanism to verify it. This utilizes blockchain-based certificate verification technology, AI-powered image recognition (for image information such as digital watermarks in the certificate), and traditional certificate chain verification technology to ensure the authenticity and validity of the server certificate. If verification fails, the client implements a secure connection rejection mechanism to prevent malicious connections.

[0487] Key Exchange and Negotiation: If the server certificate is verified, the client and server will exchange keys and negotiate encryption algorithms using an innovative key exchange algorithm and negotiation mechanism based on quantum key distribution and multi-party secure computing. Quantum key distribution technology generates absolutely secure keys, and through multi-party secure computing, the encryption algorithm and session key are negotiated without compromising the privacy of either party. The key exchange process utilizes dynamic encryption and multi-channel transmission technologies to prevent key theft or cracking.

[0488] Client Authentication: If client authentication is required, the client sends its digital certificate to the server after key exchange. The server rigorously verifies the client certificate, using advanced biometric technologies (such as fingerprint and facial recognition, combined with biometric modules in hardware devices) and encryption verification technology to ensure the client's identity. Once verification is successful, the connection is established.

[0489] 5) Data transmission:

[0490] Data Encryption: Once a connection is established, all data transmitted between the client and server is encrypted using the agreed-upon encryption algorithm and session key. Advanced dynamic encryption algorithms, such as those based on a combination of homomorphic encryption and attribute encryption, ensure that data is encrypted into highly secure ciphertext at the sender and then rapidly decrypted at the receiver, ensuring data confidentiality during transmission.

[0491] Data Integrity Verification: The SSL / TLS protocol incorporates innovative data integrity verification technology. This technology adds a high-strength message authentication code (MAC) based on a blockchain hash algorithm and machine learning anomaly detection to the data, ensuring it has not been tampered with during transmission. Upon receiving the data, the receiver verifies the correctness of the MAC value using a fast verification algorithm and artificial intelligence technology, thereby ensuring data integrity. Machine learning algorithms are used to learn data characteristics and establish a characteristic model for normal data transmission. Any data anomalies (such as incorrect MAC values ​​or data characteristics that do not match the model) are immediately alerted and addressed.

[0492] 6) Connection closed:

[0493] When data transmission is complete or the connection is no longer needed, the client or server can initiate a connection close request. Before closing the connection, both parties implement a secure cleanup mechanism to complete data transmission and cleanup, ensuring data consistency and security. After closing the connection, related resources, such as session keys and network connections, are released. Cryptographic destruction technology, utilizing homomorphic encryption principles, encrypts the resources before destruction to prevent unauthorized use.

[0494] (3) Data reception and processing:

[0495] The communication module receives commands and data from a central server or other devices. Based on the communication protocol specifications, it utilizes innovative parsing algorithms based on deep learning and natural language processing to parse the received data and extract useful information and commands. Deep learning models learn the format and semantics of data, automatically identifying and parsing various complex data structures and command formats.

[0496] Received instructions are transmitted to the edge computing unit or other modules for processing. For example, a central server might send a device control instruction. The communication module forwards the instruction to the edge computing unit, which then controls the device's operating status based on the instruction. A secure instruction transmission and verification mechanism based on blockchain smart contracts and encrypted transmission ensures the accuracy and security of the instructions. Blockchain smart contracts are used to record the transmission and execution of instructions, making them traceable and tamper-proof. Encrypted transmission ensures the confidentiality and integrity of instructions during transmission.

[0497] The implementation method of a data acquisition terminal provided in embodiments of the present invention uses edge distributed computing to push data processing to edge devices, thus creating intelligent data acquisition terminals. Edge computing nodes can rapidly process pre-processed data locally, extracting key features and performing fault diagnosis and prediction without waiting for data to be transmitted to a central server. This allows for immediate decision support for equipment maintenance and management. For example, in high-speed machining centers and other equipment, device data can be analyzed within microseconds to promptly identify potential faults and avoid production interruptions. The data acquisition terminal integrates a wide variety of intelligent data acquisition interfaces, including standard device data interfaces, Modbus TCP / RTU interfaces, and sensor interfaces, to meet the data acquisition needs of diverse equipment types. This eliminates the need for custom development for each device, improving device compatibility and versatility. For sensors without transmitters, transmitters can be flexibly added to convert their output signals into standard electrical signals, further expanding the scope of data collection. Edge computing nodes utilize advanced data analysis algorithms and machine learning techniques to conduct in-depth analysis of equipment data. They can automatically identify patterns and trends in the data and build highly accurate fault diagnosis and prediction models. For example, deep learning of historical data on equipment operating parameters can accurately predict the timing, type, and severity of future equipment failures. This allows for proactive preventive measures, reduces equipment failure rates, and improves equipment reliability and production efficiency. Clearly define equipment performance indicators, such as machining accuracy, production efficiency, and energy consumption. Real-time collection of equipment operating parameters and status data allows for intelligent performance evaluation. Comparing current performance with historical performance, analyzing trends, and identifying the causes and issues of performance degradation provide a scientific basis for equipment optimization and improvement.

[0498] Furthermore, the innovative MQTT+SSL protocol is used for data transmission, ensuring the security and integrity of data during transmission. The SSL protocol encrypts communication data to prevent data theft or tampering. The lightweight nature of the MQTT protocol and its publish / subscribe model make data transmission more efficient and flexible. When establishing a connection between the edge computing unit and the central server, strict digital certificate authentication is used to prevent unauthorized devices from accessing the network. Edge computing nodes feature local storage and support resumable data transmission. Data encryption and access control technologies ensure the security of locally stored data. Only authorized users can access locally stored data, preventing data leakage and tampering.

[0499] The edge distributed computing-based data acquisition terminal provided in the embodiments of the present invention leverages advanced data analysis algorithms and machine learning technologies to enable real-time analysis and processing of equipment data, encompassing functions such as fault diagnosis, prediction, and performance evaluation. This helps enterprises promptly identify equipment problems, implement effective maintenance measures, and improve equipment reliability and production efficiency. It also provides a scientific basis for equipment optimization and improvement, driving intelligent equipment management forward. Accurate fault diagnosis and prediction enable enterprises to proactively identify potential equipment failures and implement targeted maintenance measures to avoid losses caused by production interruptions. Through equipment performance evaluation and optimization, equipment lifespan is extended and maintenance costs are reduced. This provides enterprises with timely equipment maintenance decision support and reduces maintenance risks. Real-time data collection and analysis allows enterprises to monitor equipment operating status and performance issues at all times, enabling timely adjustments and optimizations to improve equipment production efficiency. By deeply analyzing equipment data, this provides a scientific basis for production scheduling and management, optimizing production processes and further improving production efficiency.

[0500] The data acquisition terminal in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0501] The embodiment of the present invention also provides a computer device having the above Figure 1 and Figure 2 The data acquisition terminal shown.

[0502] See also Figure 15 , Figure 15 is a structural diagram of a computer device provided by an optional embodiment of the present invention, such as Figure 15 As shown, the computer device includes: one or more processors 10, memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components utilize different buses to communicate with each other and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in the memory or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Equally, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 15 A processor 10 is taken as an example.

[0503] The processor 10 may be a central processing unit, a network processor, or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof.

[0504] The memory 20 stores instructions that can be executed by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0505] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created based on the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0506] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0507] The computer device further includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Figure 15 The bus connection is taken as an example.

[0508] The input device 30 can receive input digital or character information and generate key signal input related to user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a trackpad, a touch pad, an indicator stick, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 can include a display device, an auxiliary lighting device (e.g., an LED), and a tactile feedback device (e.g., a vibration motor). The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display, and a plasma display. In some optional embodiments, the display device can be a touch screen.

[0509] The embodiment of the present invention also provides a computer-readable storage medium. The above-mentioned method according to the embodiment of the present invention can be implemented in hardware, firmware, or implemented as a computer code that can be recorded in a storage medium, or implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the method shown in the above embodiment is implemented.

[0510] A portion of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the form in which the computer program instruction exists in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc. Accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium that can be accessed by the computer.

[0511] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A data acquisition terminal, characterized in that: The data acquisition terminal includes: a data acquisition interface and an edge computing node; The data acquisition interface is used to collect multi-source operating status data of the collected device from multiple signal sources; The edge computing node is used to receive the multi-source operating status data collected by the data acquisition interface, perform data analysis, fusion and edge computing on the multi-source operating status data, and perform fault diagnosis prediction and equipment performance analysis on the collected equipment based on the edge computing results.

2. The data acquisition terminal according to claim 1, characterized in that: The data acquisition interface includes: a standard equipment data interface, an interface based on Modbus TCP, an interface based on Modbus RTU, and an analog and / or switch quantity interface; The standard device data interface is used to connect devices or sensors with a Fanuc data interface and / or a non-Fanuc data interface; The Modbus TCP-based interface and the Modbus RTU-based interface are both used to connect devices or sensors that provide RS485 interfaces or RS232 interfaces; The analog quantity and / or switch quantity interface is used to connect a physical quantity monitoring sensor.

3. The data acquisition terminal according to claim 1, characterized in that: The edge computing node is a Linux-based development board, and the edge computing node includes: The data analysis and fusion unit is used to analyze the multi-source operation status data, obtain multiple data features, and fuse the multiple data features based on the data fusion algorithm; The real-time analysis unit is used to perform edge computing on the data features after data fusion based on a machine learning algorithm to obtain edge computing results, which include fault diagnosis prediction results and equipment performance analysis results.

4. The data acquisition terminal according to claim 1, characterized in that: The edge computing node also includes: A local storage unit, used to store multi-source operating status data according to a preset time period; The encapsulation and forwarding unit is used to encrypt and authenticate the multi-source operation status data based on the data encryption and encapsulation method, and then encapsulate and forward it.

5. The data acquisition terminal according to claim 1, characterized in that: The data acquisition terminal also includes an acquisition preprocessing module, which is connected between the data acquisition interface and the edge computing node. The acquisition preprocessing module is used to preprocess the multi-source operating status data by performing data cleaning, noise removal and data normalization.

6. The data acquisition terminal according to claim 5, characterized in that: The data acquisition terminal also includes a communication module and a power supply module; The communication module is respectively connected to the data acquisition interface, the acquisition preprocessing module, the edge computing node and the power supply module; The communication module is used to provide wired or wireless communication for data transmission between the data acquisition interface and the acquisition preprocessing module, data transmission between the acquisition preprocessing module and the edge computing node, and data transmission between the acquisition preprocessing module and the power module; The power supply module is used to output various voltage levels and automatically adjust the power output according to the working status of the data acquisition terminal; The acquisition and pre-processing module is also used to monitor the operating status and temperature, voltage and current parameters of the power module.

7. A method for implementing a data acquisition terminal, characterized in that: Applied to the data acquisition terminal according to any one of claims 1 to 6, the method comprises: Collect multi-source operating status data of the collected equipment from multiple signal sources; Receive the multi-source operating status data, perform data analysis and fusion and edge computing on the multi-source operating status data, and perform fault diagnosis prediction and equipment performance analysis on the collected equipment based on the edge computing results.

8. The method according to claim 7, characterized in that Perform data analysis, fusion, and edge computing on the multi-source operating status data, and perform fault diagnosis and prediction and equipment performance analysis on the collected equipment based on the edge computing results, including: Perform data analysis on multi-source operating status data to obtain multiple data features, and then fuse the multiple data features based on the data fusion algorithm; Edge computing is performed on the data features after data fusion based on a machine learning algorithm to obtain edge computing results, which include fault diagnosis prediction results and equipment performance analysis results.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, computer instructions are stored in the memory, and the processor executes the implementation method of the data acquisition terminal described in claim 7 or 8 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the implementation method of the data acquisition terminal according to claim 7 or 8.

Images